Redundancy has been popular technique to design a fault tolerance system. As the hardware in a digital system are getting lower in price and software redundancy will consume more time and resources, hardware redundancy is useful mechanism to improve the reliability of the system. While an active hardware redundancy is used to detect the faults in the hardware component and recover the system from those faulty components, passive hardware redundancy is basically used to mask the faults in the hardware module. Triple Modular Redundancy is suitable example of hardware redundancy technique where three identical functional components are used and voting mechanism will take majority of inputs as the output of the system. However, a single point failure in the TMR will impede to improve the reliability of the system. This report will discuss on different mechanisms to improve the voter circuit of triple modular redundancy. A report presents a detail analysis on different techniques and comparison among these technique on basis cost and complexity to implement in triple modular redundancy to improve voter circuit.
A failure of any components in a critical system could lead to catastrophic events like a human or economic loss. The divergent of a result of a non-critical system from its expected output will have little or no effect on the throughput of the system compared to that of a critical system . So, it is imperative to ensure that every component is available at all the time and performing as it is expected and produce the desired output. With an implementation of fault tolerance mechanism in a critical system, there is an increase in the availability and reliability of a system’s components. A fault tolerant system is designed in such a way that it can detect the failures of components and isolate the defect components from the existing system and produce the desired output without interrupting the flow of the system. A fault tolerance system can be implemented through hardware redundancy and software redundancy. As the hardware redundancy is comparatively lower than software redundancy, it has become a popular technique to design a fault tolerance system .
A triple modular redundancy is one of the techniques to build fault tolerance system as it increases the reliability of the system through implementation of two out of three voting mechanism technique . The voting mechanism has played a significant role in triple modular redundancy technique as it will decide an actual output out of three components. The majority outputs from the three components will be the actual output of the voting circuit. Despite its ability to generate consensual output from the three modules, the presence of single point failure i.e. the failure of the voting circuit itself could lead to inconsistency in the output the overall systems. This report includes a thorough description on fault tolerance, hardware redundancy, and triple modular redundancy technique. Furthermore, the report primarily focused on a mechanism to avoid the single point failure i.e. voting mechanism in the triple modular mechanism technique.
The concept of fault, errors and failure plays crucial role while improving the dependability of the system. Most of the times, these three terms are used interchangeably in fault tolerance. However, each term has their own meaning and successful implementation of fault tolerance design must handle each of these terms.
It can be defined as the defect or flaws in the hardware or software component of the system. If it is hardware component, fault is considered as the defects whereas if it is software component, it is considered as the flaws in the software system. Faults are basically considered as the root of the failure of the systems, so, it is necessary to avoid any sort of the faults while integrating hardware or software of the system. On basis of the duration, fault can be permanent, transient or intermittent fault. While permanent fault occurs due to failure of components, physical damage of hardware or design and do not fix by themselves along with time, transient faults occur for short period of time and do not re-occur continuously. Meanwhile, intermittent fault moves in between fault and fault free operations in the system. Both transient and intermittent faults are occurred frequently compared to that of permanent faults in the system .
Meanwhile, errors occur when output from the system is deviated from the expected output. So, when there is fault in the hardware components or software components, there will be inconsistency in the result from the components. Basically, error indicates the state that is invalid to the system specification and should not exist [5,6].
It can be defined as the incapacity of the system to produce desired results as specified in the system specification. When error occurs, a system will be in undesirable state and the result from the system will be different from the expected one. The system behavior will be changed which is considered as the failure of the system [5,6].
Redundancy is one of the efficient techniques to achieve fault tolerance mechanism in the digital system. As its name suggest, redundancy is basically an addition of hardware and software resources, information or time in a system more than it required to perform their normal operation. There are four redundancy techniques and they are:
- Software Redundancy
- Time Redundancy
- Information Redundancy
- Hardware Redundancy
While the physical defects in the hardware components can barely re-occurred once it is discovered and repair, fixing bugs in the software programs will create greater chance to create other errors in the coding. Despite the presence of different software development process, there are chances to develop error prone software due to novice developers, lack of testing or inadequate time and money for the full phase development of the software. N-version programming is one of the software fault-tolerance technique where a program writes for N times and execute in parallel to take majority output as a final output of the program. While this technique is an effective to mask the errors in the program, it will be costly and difficult to maintain all the version of the code. Likewise, a use of watchdog timers and timeouts, time redundancy for program or self-checking in the software allowed to detect the faults in the software. When a software runs for multiple times and compared the results, it is possible to identify the faults or bugs in the program.
The concept of time redundancy is to run the program multiples times in the presence of same hardware configuration and compare the produce results. It reduces the expense on expensive hardware addition and also avoid parallel execution of the programs. Since a program can be run multiple times in the same modules and compare the results to identify errors or faults, it will efficient compared hardware and software redundancy technique. Besides that, it is suitable for transient or intermittent faults as a frequent run of programs in same module will help to get majority output as the actual output of the system. However, one of the considerable disadvantages of time redundancy is the requirement of large amount of time to identify the faults and defects in the system compared to hardware and software system .
Here, there will be an addition of extra information along with data to ensure to integrity of the information changes during a storage or transmission. An error-detecting codes and correcting codes, and self-checking circuits are popular mechanism for information redundancy. Parity code is widely used for error-detection in the memory of the computer system. A parity bit is generated by a parity generator and data is encoding through computation of its parity. When there is changes in the computed parity bit with stored parity bit, there is an indication of data changes and error signal is sent to the processor of invalid memory data. On the other hand, data which are encoded with error-correcting codes contains both errors and adequate redundancy to recover the desired data. Meanwhile, self-checking circuit will produce valid output word when there is valid input and when there is existence of fault, it will produce invalid output code to detect the fault in the system .
Hardware redundancy can be achieved through the addition of extra hardware to the system. As the hardware components are getting cheaper with advancement of technology, it can be considered as suitable mechanism to achieve reliability in the system. Apart from that, it also does not require continue observation and will not take more time to identify and mask the error compared to other redundancy techniques. For example, an addition of processors, data or memory buses, power or even memories can easily achieve hardware redundancy. There are commonly three approaches to obtain hardware redundancy techniques and they are as follow
- Passive Redundancy
- Active Redundancy
- Hybrid Redundancy
Passive redundancy is responsible for hiding and covering the faults in the hardware components rather than detecting those faults. It will produce the result based on polling mechanism and provide the correct output from the system despite the existing of faults in the components. When there are multiples faults than polling circuit can cover, then it cannot hide the faults and failure of the system is imminent. Triple modular redundancy and N modular redundancy are the suitable example of fault tasking technique through use of redundant hardware in the system .
Figure 2 N-modular hardware Redundancy
Similarly, active hardware redundancy is used to detect the faults in the components and recover from those faulty components. There are use of different techniques for fault detections and computation with duplication is one of the techniques where two duplicate modules execute identical computation in parallel and with use of comparator results are compared. If the results are not equal between two modules, an error signal will be produced.
After the detection of faults, no more activity is carried out until the system recover from the fault.
Figure 3 Comparison with Duplication
Meanwhile, in standby Sparing technique, a single module out of N modules will operate and responsible for the output of the system whereas remaining n-1 modules will remain as a spare. When there is failure of working module, spares will be switched and start to operate. Therefore, a standby sparing system having N modules will tolerate N-1 faults in the modules. Furthermore, Pair and spare technique combine both comparison technique and sparing technique to improve fault detects and recover. There is a use of two modules in parallel and when there is error after comparison, it will identify defected module and replaced it will spare modules. Apart from these techniques, timer mechanism can be used for error detection [10,11].
Besides active and passive redundancy techniques, hybrid redundancy technique, which includes features from both earlier techniques, is another hardware redundancy technique to identify faults in the components and recover from those faults. One of the approaches of hybrid redundancy technique is duplex-triplex architecture where two duplication with comparison technique along with TMR is used to mask the errors, detect the faults and remove those faults from the system to produce desirable output. One of the major disadvantages of hybrid redundancy is that the methods which are used to implement it are costly .
Triple Modular Redundancy
The concept of Triple Modular Redundancy is mainly based on redundancy of hardware components of the system. It is highly used to make system more reliable and continue to perform their operations against the soft errors. When an error occurs in sequential circuits, which indicates to the different storage in the system such as registers, memories, flip-flips and counters, there will be a change in the saved state in different storage and lead to the execution of the program different from the expected one. To minimize effects of soft errors, TMR is designed in microprocessors so that errors will not halt the flow of the program.
Figure 4 Triple Modular Redundancy Technique
The generalization of TMR technique in N modular redundancy technique. While there is presence of n modules in the N modular redundancy technique, TMR will have 3 identical functional hardware units to perform the operation. The concept of the TMR design is to use three duplicate modules which will take same input. All the operation within the module is same for all three of those modules, so, the use of input data is same of all three cases. On the top of the identical modules, there is apply of voting mechanism to get majority output as the actual output of the system. So, voter circuit will take all three outputs of the hardware unit as the input to the unit and majority of input will consider as the actual output of the system. For example, consider there are three hardware modules A, B and C in the system. All three modules are functionally identical, and all are supplied with 0 input. Among three modules, two produce output result as 1 whereas one module produces 0 as output. When all three outputs from module enter into voter Logic, E, it will perform some calculation to decide majority output from the modules, which in this case will be 1 and considered 1 as the output of the component. Since inputs in the voter logic are binary and number of inputs are in odd numbers, it will be easier to calculate majority inputs in the voter machine .
The basic purpose of Triple Modular Redundancy is to mask the errors exist in the functional unit of the system. So, it can be considered as the passive hardware redundancy technique to achieve the reliability of the system. The majority logic gate also known as voting logic is composed of simple AND-OR circuit. Consider a, b and c are three inputs from the modules to the logic gate, then it can be defined as ab V bc V ac. Furthermore, mathematically, the reliability of any system, R, can be calculated as the sum of the probabilities of success and fail events i.e.
R(system) = Rm3 + 3RM2(1- RM) = 3RM2 – 2 RM3
Here, the basic assumptions are made, and they are voting circuit does not fail, failures of three modules are independent from each other, system will not fail if none of three modules fails or exactly one module fails at a time. However, the major disadvantage of triple modular redundancy technique is its single point failure. While there can be made an assumption that voter circuit never fails but when it does, the reliability of whole system will be compromised. When all three modules are working perfectly, the failure of voter circuit can result different output from the system rather than actual correct output produce by the functional unit .
The purpose of fault tolerance system is to avoid a failure of the overall system in spite of existing faults in the different components. Redundancy has been a significant technique to assure fault tolerance design in the digital system. Among several redundancy techniques, hardware redundancy is constantly used to improve the reliability of the digital systems. While active hardware redundancy is useful to detect faults and recovery, passive hardware redundancy is useful to hide faults in hardware components. TMR is a passive hardware redundancy technique where faults are hidden, and only correct data are passed as output from the system .
V.M et al (2005) had discussed on the alternative technique to triple modular redundancy technique called Reduced Triple Modular Redundancy. It basically operates on lookup-table which is achieved after technology mapping stage. Rather than using three identical entire design, they preferred for the triplicating first and third category of the lookup table and there is a use of tri buffer-based majority voters to get the output from the system. The objective of their research is to handle single event failure, which mainly occurred to due excess charged induced from the radiation. As a result, it is highly influential to change the internal state of the data, mostly on memory elements like routing configuration bits and look up table entries. So, such errors could lead to change in bits in memory or table entries and completely produce unexpected output from the system. With the help of spare flip-flops of CLBs in combinational circuits, a number of insensitive LUTs are duplicated. They described a major disadvantage of using Triple Modular Redundancy technique is an excessive area overhead. Their proposed technique, RTMR, just required 99.61% of additional number of LUTs, which is almost 100% less that of LUTs required in normal TMR. Despite the use of a smaller number of additional LUTs in RTMR, it was able to provide high level of SEU immunity .
Another proposed approach to improve Triple modular redundancy is cascaded TMRs. They are mainly used in areas like pipeline process, Poly-Si TFT. Since there is use of multiple stages of TMR design to build the cascaded TMR, it is also known as Multi-stage TMR. The structure is designed in a such way that one output from the TMR will be the three inputs for the another TMR. It can be considered reliable if at least two inputs from each stage is provide correct value and end voter produce the accurate output. With use of multiple voters, it can impede the single point failure exist in the original Triple modular redundancy technique. While an original TMR has single voter, it cannot provide reliable result on the failure of voting system whereas cascaded triple modular redundancy technique can provide reliable result despite failure of voter.
Figure 1 Classical Cascaded Single Voter TMR Module
While original cascaded TMR consists of single point failure one voter, new proposed technique consists of multiple voter per stage. When there is use of single voter in the TMR, there are strict rule to achieve the reliability of the system. Meanwhile, when there is use of multiple voters in each stage of the TMR, it will increase the cost of designing and ultimately violate the cost-effective design. So, Yi, Chung and Kim (2015) proposed new cost-effective cascaded Triple modular redundancy which will improve the cost effectiveness of the design and also ensure there is loosen in the rules to bring the reliability to the system .
Meanwhile, the concept of TMR is to mask the faulty module in the system. When there is one faulty module in the system, it can continue to work and provide desired result. In order to execute the same task on the same hardware after the detection of TMR failure or replace the hardware, reload and restart, Shin and Kim (1994) proposed an adaptive recovery method by optimally choosing either RSHW or RHWR on basis of their cost effectiveness. They used Bayes theorem to update collect information of each state in the TMR system after each voting result. When there is TMR failure, the expected cost of RSHW will calculated on basis of all the likelihoods and those results are compared with RHWR. After continuing increase in the number of unsuccessful RSHWs, permanent TMR failure will be increased and eventually there is increase in the cost of RSHW. As more than 90% faults in the system are known as transient faults or non-permanent faults and around 2% faults are permanent faults, a simple re-execution of the task will be an effective process to recover from most of the TMR failures. Such re-execution of task will help to eliminate hardware cost arise due to replacement of large chunk of hardware failures, reduce recovery time arise through the replacement of hardware, system configuration and system restart. The adaptive method of RSHW calculate the state with all possible likelihoods state in the system and later gauge the RSHW or RHWR based on their expected cost when the system reaches in one of the estimated states. When the number of unsuccessful RSHWs exceeds the maximum number of RSHWs allowable or estimated cost of RSHW exceed that of RSHW, then RHWR will be invoked .
Furthermore, Patooghy et.al (2006) discussed on distributed voting mechanism to overcome the existing problem of the single point failure in the Triple Modular Redundancy. Their ideas utilized the use of time redundancy and disagreement detector features to overcome the issue. A method is experimented with vertex2Pro and Vertex4 Xilinx FPGA to demonstrate the reliability and improve mean time failure of a TM system. They highlighted despite the continue research to improve the voting circuit in the TMR system, many of those works are based on certain level of assumptions which makes existing problem simple but unrealistic. One common example of such assumption is many of researches are based on neglecting the single point failure to improve the reliability of the system. Their proposed method would mask the permanent and transient faults which occurred in the voter of a TMR system with help of time redundancy, Triple modular redundancy with disagreement detector and use of n spares voter in the system. The number of spare voters used in the system will help to decide how many faults it can cover in the system. One of the benefits of their proposed method is that distribution of decision between voter and disagreement detector. However, one of the disadvantages of their proposed method is certain level of delay to produce the output due to comparison between distributed decision maker .
The use of word voter over conventional voter has their own advantage to ensure the data integrity. Mitra and McCluskey (2000) has discussed on word voter and presented their implementation with compare to bit by bit voting design in conventional voter system. They had compared area and delay overhead of word-voter design with bit by bit voter. The use of word-voter will have to improve the capabilities of Triple modular system to protect against common-mode and multiple module failures. There is necessary of extra hardware to build the word voter which is an order of one 2-input logic gate and three XNOR gates for each output of the system. Meanwhile, the word voter design can be modified to guarantee that data outputs without the error signal in absence of extra gate.
Figure 2 TMR system with Word Voter
Figure 3 voting mechanism in TMR
Here, the above to figure explaining the voting mechanism for both bit by bit voting mechanism and word voting mechanism. While bit by bit voting mechanism will considered each bit of the input to find the errors in the faulty module, word-wise voting mechanism will take both of the inputs and compared to find the errors in the module  .
While TMR supports to improve the reliability of systems, it’s single point failure i.e. voting logic failure could cause the failure of an overall system. Sadeghi, Soltan, and Khayyambashi  has argued that despite a simple structure of TMR model, an existence of voter logic as a weaker point will lead to a failure of the overall system. When there is a fault in a single hardware component, a voter can mask those faults by generating correct output from the system. However, if there are any faults on a voter, it will produce an inconsistent and incorrect output which can lead to the failure of a system. Therefore, they have proposed to use three voters instead of one voter so that the probability of voter logic failure will be reduced and a failure of one voter can be covered by other two voters in the system.
As the reliability of a voter must be high to improve the overall reliability of the system, an assumption of voter always produces a correct result or remain fault free weaken to produce fault tolerance design. Moslem and Václav  have proposed novel fault tolerant voter circuit design using wired-logic to make voter itself to tolerant faults and support the reliability of a system. An open drain CMOS NAND gate, which comprises only 2 series-connected n-channels are used to make fault-tolerant voter. Furthermore, novel voter circuit design can be expanded for N- modular redundancy system where N is odd integer and greater than or equal to five.
While there are different methods which are proposed to improve the reliability of systems and make fault tolerant voter circuit, some proposed methods increased the complexity with use of multiple multiplexer and priority encoders. Meanwhile, other proposed methods are lack of inextensibility due to which such module will only support on particular models rather than generalized model. A novel fault tolerant voter circuit has reduced the complexity in its design and further support for generalized hardware redundancy methods.
Triple modular redundancy method has been useful technique to improve the reliability of the system. While TMR has ability to mask the fault exist in of the module, it will not able to minimize the fault exist in the voter circuit. It is also called single point failure in the TMR system and failure of voter circuit can compromise the reliability of the system. This report has performed primary research on improving the voter circuit of the Triple modular redundancy system. Meanwhile, the report also performs the secondary research on improving the reliability of triple modular redundancy system with consideration of TMR failure, and so on.
The problem statement of the report clearly states that voter circuit in the triple modular system fails can lead to the failure of the system. Voter circuit is used to produce an output from the majority of the input to the voter. Since there is use of three identical modules in the TMR with providing same input on each of those modules, voter circuit can receive either same input from all the modules or can receive at least two different input from three modules. The report has identified different assumptions that are made while designing the triple modular redundancy and elaborate and eliminated those assumption to identify actual technique that would help to improve the voter circuit of the system. Mainly, research finds some of the assumptions common on different implementation technique and they are as follow
- Voter circuit never fails
- There will be no faults on two modules at once, which means only one module can have fault to implement TMR design
Both of these assumptions have made significant role in designing simple Triple Modular Redundancy system.
Since a research is a qualitative research, it has described different terminology associated with TMR technique. The distinguish between fault, error and failures in digital system followed by different redundancy techniques such as hardware, software, time and information redundancy technique helps to enlist the advantage and disadvantage of each redundancy technique. The research illustrated that hardware redundancy is more common practice to build reliability in digital system compared to that on software redundancy. As hardware devices are cheaper along with advancement in technology, it can be useful to design fault tolerance system. On the other hand, software redundancy required multiple times run of same program in parallel, which will consume both resources and time to find the faults in the program and fix those bugs. Furthermore, research also identifies time redundancy can be useful to avoid the use of additional hardware and need of parallel run of the program. However, since it executes multiple times of the program to identify the bugs in the program, it will delay overall execution of the program.
Meanwhile, to improve the voter circuit, this report has measured the cost effectiveness and complexity of proposed design to improve reliability of the voter circuit. A literature review from different papers were observed as a part of the survey to identify the potential design to improve the voter circuit of Triple modular redundancy system. In addition, research report has compared the theoretical knowledge of different papers rather than collecting sample data and applying numerical formula to compare the result. Based on results from different papers, benefits and drawbacks of each design is identified and classified to choose appropriate design method to resolve an issue associated with the single point failure. While the paper has tried to minimize unnecessary assumptions but has not completely avoid all the assumptions exist while designing the system.
The report has reviewed several research papers related to reliability of triple modular redundancy. It is observed that triple modular redundancy has simpler structure to design and effectively handle the fault masking of hardware components. The observation of results from different papers derived that the assumptions such as no failure of two hardware components at once and simply voting logic will never fail has weaken the reliability of triple modular redundancy technique. Meanwhile, a research also finds despite many of these proposed techniques are viable to improve voter, a complexity, inextensibility, and dependency on the other components make difficult to implement. By comparing above factors, a novel fault-tolerant voter circuit design is suitable to improve voter logic as it overcomes the complexity, inextensibility, and dependency and able to provide reliability to the TMR system. It has reduced the complexity required to increase the reliability of the voter circuit as there is only use of twelve transistors and one pull-up resistor for a voter. Furthermore, the dependency due to the network on one gate in the voter to produce the final output is also reduced.
Figure 1 Reliability Comparison between classic voter and novel voter
With use of open-drain NAND gates, which consists of only 2 series-connected n-channels, a simple and reliable voter circuit can be generated.
Conclusion and Future Work
Hardware redundancy is useful to improve the reliability of the system. Although the structure of triple modular redundancy is simple and cost effective, a single point failure i.e. voter circuit could cause the failure of the overall system. Meanwhile, different techniques were proposed to improve the reliability of the triple modular redundancy, however, many of these techniques were based on assumptions that no two modules will fail at the same time and voter circuit will never fails. The report has performed detail analysis on some techniques which will be useful to improve the voter circuit in the system.
This research remains focused on improving voter circuit of the triple modular redundancy. The report studies several literature review papers to analysis the technique to improve the voter mechanism so that the reliability of the triple modular redundancy will not compromise due to the single point failure in the TMR. A further detailed research and perform the analysis through collection of data will provide sheer information regarding suitable technique to improve reliability of the system. Although the report has performed cost and complexity analysis to select appropriate technique for voter, use of numerical data will further evidence to select appropriate mechanism.
 Kim, M. H., Lee, S., & Lee, K. C. (2010). Kalman predictive redundancy system for fault tolerance of safety-critical systems. IEEE Transactions on Industrial Informatics, 6(1), 46-53.
 Shooman, M. L. (2003). Reliability of computer systems and networks: fault tolerance, analysis, and design. John Wiley & Sons.
 Lyons, R. E., & Vanderkulk, W. (1962). The use of triple-modular redundancy to improve computer reliability. IBM Journal of Research and Development, 6(2), 200-209.
 Nelson, V. P. (1990). Fault-tolerant computing: Fundamental concepts. Computer, 23(7), 19-25.
 Dal Cin, M., & Hohl, W. (Eds.). (2012). Fault-Tolerant Computing Systems: Tests, Diagnosis, Fault Treatment 5th International GI/ITG/GMA Conference Nürnberg, September 25–27, 1991 Proceedings (Vol. 283). Springer Science & Business Media.
 Kaur, J., & Kinger, S. (2014). Analysis of different techniques used for fault tolerance. IJCSIT) International Journal of Computer Science and Information Technologies, 5(3), 4086-4090.
 Dubrova, E. (2013). Fault-tolerant design (pp. 55-65). New York: Springer.
 Nicolaidis, M. (1999, April). Time redundancy based soft-error tolerance to rescue nanometer technologies. In vts (p. 86). IEEE.
 SADEGHI, M., SOLTAN, H., & KHAYYAMBASHI, M. (2015). The study of hardware redundancy techniques to provide a fault tolerant system. Cumhuriyet Science Journal, 36(4), 236-245.
 Oliveira, D. A., Rech, P., Quinn, H. M., Fairbanks, T. D., Monroe, L., Michalak, S. E., … & Carro, L. (2014). Modern GPUs radiation sensitivity evaluation and mitigation through duplication with comparison. IEEE Transactions on Nuclear Science, 61(6), 3115-3122.
 Dumbri, A. C., & Procyk, F. J. (1985). U.S. Patent No. 4,494,220. Washington, DC: U.S. Patent and Trademark Office.
 Lee, S., Lee, S., Lee, K.-C., & Kim, M.-H. (2017). Analytical hybrid redundancy system for the fault tolerance of advanced driver assistance systems. Proceedings of the Institution of Mechanical Engineers, Part D: Journal of Automobile Engineering, 231(12), 1660–1665. https://doi.org/10.1177/0954407016684265
 Anjankar, S. C., & Kolte, M. T. (2014). Fault Tolerant and Correction System Using Triple Modular Redundancy. International Journal of Emerging Engineering Research and Technology, 2(2), 187-191.
 Avizienis, A. (1985). The N-version approach to fault-tolerant software. IEEE Transactions on software engineering, (12), 1491-1501.
 Rahman, M., Rafique, S., & Alam, M. (2017). A Fault Tolerant Voter Circuit for Triple Modular Redundant System. Journal of Electrical and Electronic Engineering, Vol5, (5), 149-159.
 V. M. Noor Mohammad, Vikram Chandrasekhar, V. Kamakoti, “Reduced triple modular redundancy for tolerating seus in sram-based fpgas”, Proceedings of NASA International Conference on Military Applications in Programmable Logic Devices (MAPLD) September 2005, september 2005.
 Yi, H. J., Chung, T. S., & Kim, S. (2015). A Reliability Analysis of Cascaded TMR Systems. In Information Science and Applications (pp. 369-376). Springer, Berlin, Heidelberg.
 Shin, K. G., & Kim, H. (1994). A time redundancy approach to TMR failures using fault-state likelihoods. IEEE Transactions on Computers, 43(10), 1151-1162.
 Patooghy, A., Miremadi, S. G., Javadtalab, A., Fazeli, M., & Farazmand, N. (2006, September). A solution to single point of failure using voter replication and disagreement detection. In Dependable, Autonomic and Secure Computing, 2nd IEEE International Symposium on (pp. 171-176). IEEE.
 Mitra, S., & McCluskey, E. J. (2000). Word-voter: a new voter design for triple modular redundant systems. In VLSI Test Symposium, 2000. Proceedings. 18th IEEE (pp. 465-470). IEEE.
 SADEGHI, M., SOLTAN, H., & KHAYYAMBASHI, M. (2015). The study of hardware redundancy techniques to provide a fault tolerant system. Cumhuriyet Science Journal, 36(4), 236-245.
 Sheble, N. (2003). More is always better when it’s critical. INTECH-RESEARCH TRIANGLE PARK NC-, 50(10), 66-70.
 Amiri, M., & Přenosil, V. (2013). Design of a simple reliable voter for modular redundancy implementations. DISTANCE LEARNING, SIMULATION AND COMMUNICATION 2013, 8.
has been working intensively based on achieving vessel optimum trim to increase power saving and reduce greenhouse gas emissions. In this work a simplified ship model was replicated in Ansys Fluent to validate and study trim optimization. Different tests for prediction of trim optimization are presented and results are compared against each other and with literature. Results demonstrate the advantages of using Computational fluid Dynamic (CFD) analytical methods as effective design tool in marine field as it significantly reduce the time and cost of studies before depending on tests of physical models during preliminary ship design.
Drag Reduction, Trim Optimization, Numerical Simulation, Computational Fluid Dynamics, Fuel Consumption, Ansys Fluent.
In the past years the computer speed has been increased exponentially and more sophisticated RANS codes have been developed. More realistic simulations are able to be performed. All these advances have been well documented in the proceedings of several international conferences on the CFD techniques applications to ship flows. These conferences have been held every year since 1990. The most famous are those in Tokyo in 1994 , Gothenburg in 2000  and Tokyo in 2005 . Remarkable information on these application of CFD codes to naval ships and submarines can be found in recent studies shown in recent proceedings of the ‘Symposium on Naval Hydrodynamics’. This symposium is sponsored by the US Office of Naval Research (ONR) and was first held in 1956. Some examples of the advancement of current specialized CFD codes for the flow simulation around naval hulls can be found in the work of Burg et al.  reported at the 24th Symposium on Naval Hydrodynamics. It was used in Burg’s study a three-dimensional unstructured, paralyzed CFD code named U2NCLE, developed by the Computational Simulation and Design Centre at Mississippi State University to simulate the free surface flow around a fully-appended model of a US naval ship. The results found by using this code realistically captured the turbulent flow and the vortices arising from the bulbous bow and the tips of the propulsors and rudders. Also it was found that to accurately model the propulsor rotation. Near-field wave shapes were successfully simulated by a nonlinear free surface algorithm. CFDShip-Iowa code was developed at the University of Iowa’s Institute of Hydraulic Research (IIHR) under funding provided by ONR, can be used as another example of a RANS code. This code was developed to work specifically on surface-ship interface and marine propulsor flow problems. General application in naval hydrodynamics include the prediction of resistance that include friction and pressure drag prediction, wave profiles, sea keeping using a code permitting a six-degree of freedom, and manoeuvring. Wilson et al.  recently compared the performance of CFDShip-Iowa code with two commercial CFD codes. Fluent, developed by Fluent Inc. and Comet, developed by CD-Adapco, in order to predict the ship generation of waves. It was found that each of the codes has different advantages and disadvantages respecting the other one, and that each code has certain specific parameters to be considered for obtaining accurate solutions of a surface ship wave fields. It was also found that commercial solvers may present and advantages when complex geometries shapes involving appendages and propulsors are being considered because of the flexibility to use unstructured create mesh contain unstructured as well as hybrid grids. Both code, Fluent and Comet, also presented the advantages of allowing solution based grid adaption techniques to provide finer grid resolution in critical regions of the grid. It can be very useful when an interface region is considered as in this case, and hence may offer a more detailed and computationally economical way to provide accurate results of the free surface prediction problems in the vicinity of surface ship hulls. Ship trim optimization operations are considered as one of most economically wise methods for both ship performance optimization and reduction of fuel consumption. There are no requirements of shape modification or puissance upgrade in such operation. It can be achieved by preparing a proper loading plan and efficient ballasting/de-ballasting process. Operationally, Trim optimization are not considered as important as other conventional power performance test procedures. Nevertheless, it was proven that trim optimization can maintain potential savings. Also, depending on the type of ship and its operation, trim optimization could result a return on investment between one to six months. These outcomes have been proven also by Hansen and Freund in 2010 , describing depth water influence on gains. Trim optimization test procedures can be employed in towing tank model testing or in computational fluid dynamic (CFD) simulation. Hansen and Hochkirch (2013)  demonstrated possible power gains by CFD tests and gave some CFD method comparisons with tests at model and full scale. Regarding the tendency of power requirement with respect to trim, both methods were found to be in reasonable agreement. A leading consultant in trim optimization as Force Technology showed possible fuel savings of up to 15% at specific work conditions. In overall fleet operation, these savings can arrive until 2% or 3%. Extensive research campaigns have been made by Force Technology to understand physical effects to reduce propulsive power. Lemb Larsen et al. in 2012  presented an analysis of the origin factors in resistance and propulsive fields and the influence on power requirement. Even though resistance causes the most of power gain, overall performance can also be affected by variation in the propulsive coefficients, which could be considered as well.  In this paper, it was not taken in count operational constrains that should be considered in shipping, slamming or green water, strength and stability, manoeuvrability and overall safety. It was principally studied the trim optimization.
2. IMO: Ship Energy Efficiency Management Plan
Recently, energy and environment issues have been studied carefully in all around the world. The objective is to promote the energy efficiency of transportation vehicles such as ships and airplanes. Nowadays, global temperature has risen about 2°C above the pre-industrial level and it is expected that it would be cause several destruction on global scale. The only way to overcome this problem is to decrease global greenhouse gas emissions. Studying future scenarios show that if preventive actions are not taken, levels of CO2 emissions from ships would be doubled by 2050. European Union (EU) and United Nations framework convention on Climate Change (UNFCCC) are working hard to get control of global emissions as shown by Kim, Hizir, Tuan, Day and Incecik, in 2017 . International Maritime Organization (IMO) is currently controlling to set GHG commands for international shipping with industrial, operational, and market-based policy tools. It seems clearly IMO regulations concerning to CO2 emissions from shipping will remain to develop in coming years. Thus, fuel prices have ever been increasing and anticipated to continue growing in the future as demonstrated by Kim, Hizir, Tuan, Day and Incecik, in 2017 . Several operators are promoting ship’s performance should be conduct to do constant routine maintenance and control fuel consummation efficiencies. Every day offers the opportunity to optimize speed, find the safest route and make sure the ship is sailing at the best draft and trim that is tuned to keep course safely and efficiently. These efforts are compatibles directly to the goals of recent IMO Guideline on Ship Energy Efficiency Management Plans, a framework that captures the corporate commitment to energy conservation. Most important factors that should be considered for energy conservation on ships in service are as listed below: 1. Voyage Speed Optimization 2. Weather Routing – Safe Energy Efficient Route Selection 3. Hull Roughness and Its Impact on Resistance 4. Trim/Draft Optimization In this report Trim Optimization was studied carefully to find optimal points for a simplified geometry of a ship
3. Trim Optimization
Firstly, trim is defined as draught at AP and FP difference, as defined in the following expression: Trim = TA –TF In positives values trim results to the aft. Moreover, the displacement and speed keep constant when the ship is trimmed. There is not extra ballast added and power consumption varies only if resistance varies with the trim. Global goal of trim optimization is to minimize the required power at specific displacement and velocity of the ship. Physical effects of reducing propulsive power (PD) when a ship is trimmed can be caused by the hull resistance ( ) and total propulsive efficiency ( ) as shown below: PD = RT.VηD Ship speed (V) keeps constant, as said before. So, it can be concluded that the aim is to reduce the resistance and increase the total efficiency.
3.1 Reduction of Resistance
According to ITTC standards, still water ship resistance is written by: RT=1/2.ρ.V2.S.CT Ship resistance changes can be a wetted surface area (S) and total resistance coefficient ( CT) function. Both parameters have to be decreased in order to get a trim gain. Wetted surface is calculated for the ship at rest, without dynamic sinkage and trim. Variation of wetted surface are caused by trim relates to the large flat stern area and it is relatively small. It can arrive to 0.5% of the even keel wetted surface and it causes that total resistance varies because of the linear proportionality. Total resistance coefficient can be achieved by reducing all parameters mentioned before. It can be described by the following expression: CT= CR+1+k.CF0+CA Allowance coefficient ( CA) keeps constants unless for ships with big variation in the draught. Friction resistance coefficient (CF0) can vary with the Reynolds number (Re) for the flow along the hull: CF0=0.075(log10Re-2)2 Where Reis the Reynolds number defined by: Re=V.Lwlν From friction resistance coefficient and Reynolds number, it can be deduced that friction resistance coefficient is a function of the water line length ( Lwl), and this relation is inversely proportional. Although water line length can vary 5% from the even keel condition, inverse proportionality results by increase or decrease propulsive power of 0.5%. This effect comparing with overall possible savings is negligible. Form factor, ( 1+k), is often kept invariable at each draught to optimize the experimental program cost in the towing tank. In practice, there is no influence giving by this factor on the resistance changes for different trimmed conditions. Residual resistance coefficient ( CR) is often the parameter most affected by the ship trim. It was seen before that residual resistance coefficient at trimmed conditions can rise up to 150% from even keel condition values. That is reflected in power requirement changes up to 20%.
3.2 Increase of Total Propulsive Efficiency
Propulsive efficiency is the product result between the hull efficiency ( ηH), the open water propeller efficiency ( ηo) and the relative rotating efficiency ( ηrr). ηD=ηH.ηo.ηrr When the ship is trimmed none of these parameters are constant. Hull efficiency is a function of the thrust deduction ( t) and the effective wake fraction ( w). ηH=1-t1-w It is obvious that the thrust deduction decreases and effective wake fraction increases to achieve a gain trimming. Thrust deduction is a function of the propeller thrust ( T) and the hull resistance. t=T-RTT Also, it is known that hull resistance varies when the ship is trimmed and it causes that propeller thrust changes as the speed stays constant. Nevertheless, hull resistance is not constant. Thrust deduction changes as the trim changes and sometimes a peak, when propeller submergence decreases until arrive to a critical level, as it can be observed. The peaks localization depends on the dynamic sinkage and stern wave. Thrust deduction changings can produce values up to 15%. That means propulsive power changes up to 3%. However, the thrust deduction changes should be relatives to effective wake changes. Effective wake fraction is related to the ship velocity and the propeller inflow velocity (VA). w=V-VAV If the ship speed keeps constant, the effective wake fraction changes can only be related to the propeller inflow velocity. As mentioned before, the effective wake fraction increases for bow condition of trimming and decreases for stern condition of trimming. Wake fraction increasing for bow trims can catch up to 20% and the stern trims decreasing can rise up to 10%: Wake fraction differences can results in a power gain up to 2%. Propeller open water efficiency depends on the advancing ratio ( J), on the water inflow velocity of the propeller ( VA) and on the revolutions ( η). J=VAn.D Where Dis the propeller diameter. As concluded, propeller inflow velocity is affected by the trim. Since open water curve for the propeller efficiency is inclined for the actual ratio of advancing, with minor changes in the advance ratio result in a propulsive power changings. These changes can vary up to 2% of the even keel power demand. Relative rotating efficiency is described as the relation between the open water propeller torque coefficient (KQow) and the propeller torque coefficient behind the ship (KQship). ηrr=KQOWKQship It can be reach up to 2% from even keel condition and influencing the power requirement.
4. Experimental Data
There exist extensive database for CFD validation models. Some tests were done by the Resistance Committee of the 22nd International Towing Tank Conference to create this data base . The focus was on modern ship forms. Forms as tanker (KVLCC2), container ship (KCS), and surface combatant (DTMB 5415) were recommended for use. These results were shown at the workshop of Gothenburg 2000 on CFD for ship hydrodynamics  and other conferences. The bare hull trim results of this study were compared with the results of Olivieri et al.  and a combined effort between Istituto Nazionale per Studi ed Esperienze di Architettura Navale (INSEAN, Italian ship model basin ) and the Iowa Institute of Hydraulic Research (IIHR) to show experimental towing tank data and then compare to the CFD results. The CFD results used for compare appended model were the existing towing tank results performed by David Taylor Naval Ship Research and Development Centre . This model has fixed pitch shafts and struts and is designated Model 5415-1.
5. CFD Model
Computational Fluid Dynamics or CFD is a computational tool developed for studying the behaviour of the fluid flows in many different fields. It is impressive how the computer performance in this era gives us the possibility to recreate fluid flow models perfectly reals and the possibility of examining different equipment designs or compare performance under different operating conditions.
5.1 Ansys Fluent Code
Fluent code is a cell centred of finite volume general resolution for CFD problems developed by Fluent, Inc. and now marketed by ANSYS, Inc. which is a provider of engineering simulation software that acquired Fluent, Inc. in 2006. Ansys Fluent has been used by the Hydrodynamics Research Group within MPD for several years to recreate problems involving submarine related flow fields. A big difference between the simulation described in this study and those reported by the Hydrodynamics Research Group is the presence of the air/water interface in the simulation. To estimate this interface there are two main approaches, either surface fitting approaches or surface capturing approaches. Ansys Fluent implemented the surface capturing approach by using VOF (Volume of Fluid) scheme for general multiphase flow modelling. This approach involves defining a volume fraction of each phase defined throughout the domain and then convecting the volume fraction of each phase with the average fluid flow. The interface between both fluids is defined from the volume fraction function for each of phase defined before in the near of the interface. The simulation of the surface ship motion not only requires maintaining a sharp interface between the water and the air phases but requires the specification of correct boundary conditions as well at the inlet and outlet of the domain for each of the phase defined in the model. One method to do this is to impose a User Defined Function to specify the total pressure and the fluid volume profile for both fluids at both inlet and outlet conditions. Another way is to use the Open Channel Boundary Condition which allows the inflow and outflow conditions to be specified from the inflow velocity and the free surface level and then the pressure and fluid volume profile can be automatically calculated. This could be simplified the simulation. Ansys Fluent offers a package of different RANS turbulence models. These include the Sparlart-Allmaras model, the k-ε model, and the k-ω model, all of which are based on the Boussinesq approximation. Boussinesq approach assumes the calculated turbulent viscosity coefficient is isotropic. Furthermore for complex flows involving streamline curvature, swirl, rotation, and rapid changes in strain rate the Boussinesq approach is not a good approximation. For these cases Ansys Fluent provides the Reynolds Stress Model (RSM), which solves the RANS equation by solving additional transport equations for each of the individual Reynolds stresses. This means that there are five additional transport equations that are solved in all three-dimensional flows and the simulation time increases substantially. In this study the simulation described involves only straight-ahead motion at minimal angles of attack. It is expected that the flow to remain largely attached to the ship surface, hence it was employed the standard k-ε turbulence model and standard wall functions for this campaign of simulations.
The three fundamental principles govern the physical aspects of any fluid flow are the conservation of mass, the conservation of energy and Newton’s second law. These fundamental principles are expressed in mathematical equation forms, which in a general form are integral or partial differential equations. In most cases these expressions cannot be solved analytically. Computational Fluid Dynamics (CFD) uses appropriate discretized algebraic forms to replace the integral or partial derivatives in the principles equations. These discretized algebraic forms can be solved easier. The CFD solution outcome is numbers that describe the flow filed at discrete points in time and space. Navier-Stokes equations refer to the complete system of flow equations, which solves momentum, continuity and energy equations. The complete flow system equation for the solution of an unsteady, three-dimensional, compressible, viscous flow is: Continuity Equation expression: DρDt+∇.ρ.V=0 Momentum Equation expressions: ∂(ρu)∂t+∇.ρuV=-∂p∂x+∂Txx∂x+∂Tyx∂y+∂Tzx∂z+ρfx ∂(ρv)∂t+∇.ρvV=-∂p∂y+∂Txy∂x+∂Tyy∂y+∂Tzy∂z+ρfy ∂(ρw)∂t+∇.ρwV=-∂p∂z+∂Txz∂x+∂Tyz∂y+∂Tzz∂z+ρfz Energy Equation expression: ∂∂tρe+v22+∇.ρe+v22V=ρq+∂∂xk∂T∂x+∂∂yk∂T∂y+∂∂zk∂T∂z-∂up∂x-∂vp∂y∂wp∂z+∂uTxx∂x+∂uTyx∂y+∂uTzx∂z+∂vTxy∂x+∂vTyy∂y+∂vTzy∂z+∂wTxz∂x+∂wTyz∂y+∂wTzz∂z+ρf.V Governing conservation form of equations derives from a control volume fixed in space with a fluid flow through it. Nonconservation form corresponds to the control volume moving with the fluid such that the same fluid particles are always in the same control volume. The distinction between conservation and nonconservation form was triggered by CFD development and the problem equations was more favourable to use for a given CFD application. Conservation form of the equations is more favourable for a numerical and computer programming perspective. Equation forms for an inviscid flow more known as Euler equations can be derivate from Navier-Stokes equations given above and dropping all terms corresponding to friction and thermal conduction. Navier-Stokes equations describe any fluid flow, but not all flows in the same way. The boundary conditions, and often the initial conditions, determine the particular solutions of the general governing equation for a given problem.
5.3 Discretization Method
Most of CFD codes use a finite volume method of discretization, and Ansys Fluent code, used for this study, is not an exception. As that was presented above, the governing equation or Navier-Stokes equations are composed by differential equations. It represents a problem and it is necessary transform these equation into numerical equations that contain only numbers (not expression). This process produces a numerical analogue to the derivate is named numerical discretization. The discretization is made for a simple spatial domain into finite control volumes. A control volume can decompose in many elements of a mesh and, therefore, the governing equations are solved for each element of this mesh, as shown in Figure 1. Figure 1. Volume finite method mesh element. The resulting integral equation laws would be exactly satisfied for each element and likewise for the entire domain.
Navier-Stokes equations can be solved analytically for a very small number of cases and a numerical solution is required. A numerical solution involves the governing equation discretization of motion. Once the partial differential equations are discretized it is called finite differences, but if the integral equation form is discretized it is known as finite volumes. There are some simplifying assumptions that can be made to achieve an analytical solution or to significantly reduce the computational effort demanded by the solution. This is the case when an incompressible RANS equation model is solved. Considering an incompressible flow, which is the most fluid flows assumption, the governing equation of continuity and momentum are simplified and the solution of the energy equation is no longer needed. The three velocity components are represented as a slowly varying mean velocity with a rapidly fluctuating turbulent velocity around it by the Reynolds averaging process. This method also introduced six new terms, known as Reynolds stresses. These new terms represent the increase in effective fluid velocity due to eddy turbulent existence in the flow. The turbulent models introduction serves to represent the Reynolds stresses and the underlying mean flow interactions and to close the RANS equations system. In this study the hull flow was computed using RANS equations that are becoming a standard for the numerical prediction and a viscous free surface flow analysis around ship hulls. Continuity and momentum equation, for an incompressible flow, are described by the following expression: ∇.U=0 ρU=-∇p+μ∇2U+∇.TRE+SM Where U is the averaged velocity vector, P is the averaged pressure field, μis the dynamic viscosity, SM is the momentum sources vector and TRE is the Reynold tensor stresses, computed in agreement to the K-Epsilon (k-ε) turbulent model. The Finite Volume commercial code ANSYS Fluent was used for the simulation of RANS equations on trimmed unstructured mesh.
The K-Epsilon (k-ε) model is one of the most common turbulence models. The K-Epsilon turbulent model is a two-equation model known to be an eddy viscosity mode. Eddy viscosity models use the approach of a turbulent viscosity to model the Reynold stress tensor as a mean flow quantities function. In the K-Epsilon model additional transport equations are solved for the turbulent kinetic energy k and its dissipation rate ε to enable the turbulent viscosity derivation. The model transport equation for k is derived from the exact equation, while the model transport equation for ε is obtained using physical reasoning to its mathematically exact counterpart. In the procedure of the K-Epsilon model derivation, the flow is assumed as fully turbulent, and the effects of molecular viscosity are negligible. The standard K-Epsilon model is therefore valid only for fully turbulent flows.
Eulerian multiphase model is required to create and manage the two Eulerian phases of the simulations with free surface models, where a phase has a distinct physical substance. The two phases for this models are water and air, each defined to have constant density and dynamic viscosity adjusted according to the tank average temperature. This model is not required for the models that do not use free surfaces (where the inlet fluid is water), which is gain defined to have constant density adjusted to the experiment temperatures used to validate the CFD simulation. In the Eulerian Multiphase model, the different phases are treated mathematically as interpenetrating continua. As the volume phase cannot be occupied by other phase, the concept of phasic volume fraction has to be introduced. The volume fraction parameter is assumed to be a continuous function of space and time. All volume fraction addition has to be equal to the unit. The conservation equations for each phase are derivate to obtain a package of equation with similar structure for all phases. These equations are closed by giving constitutive relationships that are found from empirical information. For the case of granular flows, the application of kinetic theory is necessary.
As already known the Volume of Fluid approach is used in combination with the RANS solver to determine the localization of the free surface. In this method this location is captured implicitly by determining the boundary between the water and the air long to the computational domain. An extra conservation variable is introduced and determines the proportion of water in the particular mesh cell with a value of one assigned for full and zero for empty. For the simulations where is no free surface, where there only on fluid, this model is no longer selected.
The simplified geometry was considered as a hydrofoil which is a lifting surfaces, or foil, which operates in water. It can be seem similar in appearance and purpose to an aerofoil used by airplanes. Figure 2. Hydrofoil geometry. The characteristics of the hydrofoil are the following:
- Hydrofoil (aerofoil adapted geometry) type: NACA0020 (Symmetrical)
- Hydrofoil Chord: 63 mm
- Hydrofoil Wing Span: 49mm
The geometry for this study was created in Cartesian coordinates. The geometry consisted of a simplified ship with six surface boundaries: inlet (for inflow), outlet (for out pressure), wall flow (wall), airflow (wall), symmetry1 (symmetry) and symetry2 (symmetry). The pre-processing of the surface and initial meshing was done in Ansys ICEMCFD software and subsequently the surface was imported into CFD code Ansys Fluent. The mesh contains approximately 38.000 cells. A structured mesh was used to this study. The ship surface was first meshed using quadrilateral elements. A volume forward of the bow was then created by projecting the outline of the bow surface towards the inlet. The mesh of this volume was created from the projection surface mesh on the bow matching portion. Once the surface mesh is created and physics models are selected, the next is to create the volume mesh. Volume controls are utilized to make the mesh more efficient and more effective as well. A volumetric control is used to specify the mesh properties as density for the surfaces and volume type meshes for the mesh generation. These controls work with volume shapes in the software. The volume shapes are a pretty well approximation of a geometry figure that can be used to specify a volumetric control for the surface or volume mesh refinement or coarsening during the meshing process. Volume shapes are created to achieve more computationally demanding and computationally important spaces of the mesh. The space around the bow, the stern and the free surface, and the space around the free surface as well, up to the height of the generated waves. The global goal of these volumetric controls, these spaces covered by the volume shapes, is specifically to have a more refined mesh. The driving factor which is in charge to the mesh refinement process and to maintain a balance between having satisfactory results and resting the computational cost as low as possible. The geometry was meshed with tetrahedral cells of defined global size, as shown in Figure 1 The mesh is finer at the more complex areas of the ship surface in order to capture the extra details and to more accurately represent them. The dimensions of the block that defines the volume mesh region. Figure 3. Mesh plot The dimension of the entire domain was chose with regard to accuracy of the results. The dimension was limited as much as possible because of the computational cost or simulation time. The length dimension behind the ship stem is longer than that forward from the bow to get more details on the waves generated by the ship. The dimensions of the block for the simulations with a free surface were initially similar in size to the ones of the simulations without a free surface. Nevertheless, as pressure concentrations were found at the ship boundaries, they were gradually increased to better represent the fluid flow and improve the accuracy of the results.
For the mesh evaluation two scalar and dimensionless parameters were used to evaluate the generated mesh. These parameters were wall y+ and the convective Courant number. The convective Courant number is only used in implicit unsteady models, thus this parameter was used to validate the model. Boundary layer needs a high mesh resolution in the near-wall region. The normalized wall distance parameter y+ was used to check the quality of the mesh near the walls and within the boundary layer. This parameter is defined as: y+=Twρ.yν Where Twis the shear stress at the wall, ρ is the local density, y is the normal distance of the cell centroid from the wall and ν is the local kinematic viscosity. Potential errors appear with large values of y+. If it is used a high y+ wall treatment, it is generally prudent to have y+ values between 30 and 50. Moreover, there are some cells that will inevitably have a small value of y+. Most values of y+ below 100 are considered as reasonable. The lowest y+ wall treatment requires the entire mesh to have values of y+ approximately of 1 or less. In this study, the all y+ wall treatment is used because it is the most general and all values of y+ are fixed to be below 100. Convective Courant number is defined as: Convective Courant number = Vdtdx That means to evaluate the mesh with the chosen time step. Convective Courant number depends on the velocity V, the time step dt, and the interval length or length of the cells dx. This ratio of the time step and the time required for a fluid particle to travel the cell length with its local speed. For each cell this parameter is calculated and it gives an indication of how fast the fluid is going through the computational cells defined before. As it can be deduced for a finer mesh drives the Courant number at higher values, a smaller time step drives it at lower values, and a higher velocity drives it up. Implicit solvers are usually stable at maximum values in the range 10-100 locally, but with a mean value of about 1. The Courant-Friedrich-Lewy condition states that the Courant number should be less than or equal to unity. Generally, this parameter arrives to values less than 1 and it is expected to give models that run faster and with greater stability.
The work conditions of the simulation were the same that those in the Gothenburg 2000 workshop. The model attitude respecting to the coordinate axes was set according to the experimentally measured sinkage and trim values before creating the mesh. The coordinate system origin is located at the middle ship intersection of the water free surface and the centre plane. The open channel boundary condition was used to specify the inlet and outlet boundary condition. As has already said before, the boundary condition gives the particular solution of the general governing equation to any flow. Furthermore, these numerical solutions of the Navier-Stokes equations must give a compelling numerical representation of the proper boundary conditions. It was considered a no-slip wall boundary condition at all the walls (wall-flow and airflow and. The no-slip wall boundary condition represents the proper physical condition for a viscous flow, where the relative velocity between the boundary surface and the fluid immediately at the surface is assumed to be zero. The velocity of the flow at the surface is zero if the surface is stationary with the flow moving past it as it was defined in this case. At the inlet, it was prescribed a constant velocity that corresponds to the Froude number at which the simulation was run. The direction of the velocity that means the direction, at which the flow moves, is that of the x-axis direction. To understand, it moves perpendicular to the inlet boundary surface and toward the outlet. This inlet velocity condition is suitable for an incompressible flow. It is often utilized in combination with a pressure outlet boundary condition at the outlet flow, as it was employed in this study. The pressure outlet boundary condition is the flow outlet boundary at the outlet pressure defined. The pressure was specified to be the hydrostatic pressure of the flow with the reference pressure is the atmospheric pressure at sea level. Also, for the symmetry boundary conditions it was assumed a symmetry condition. A symmetry plane boundary condition is better used for the physical geometry with an interest and the expected pattern of the flow has mirror symmetry. A surface is defined as a symmetry plane boundary condition if it is the imaginary plane of symmetry in a simulation that would be physically symmetrical if modelled in its entirety. This solution for a symmetry plane boundary is identical to a solution that would be obtained if the mesh was replayed about the symmetry plane but in the other half of the domain. In this study, the simulations presented have an imaginary plane of symmetry where this boundary case condition was appropriate to be used. A zero-shear slip wall in viscous flows can be also modelled by a symmetry boundary condition. It was found that this condition works well and it was also used at the side, bottom and top boundaries of the simulations without a free surface. For these boundaries in the simulation with a free surface the velocity inlet boundary condition provided better performance results. The velocity imposed was the same in magnitude and direction as at inlet boundary condition. According to the real model physic, the numerical model develops a laminar regime all along the pipe and a laminar model was selected to solve the governing equations for this case. Also, the fluid was supposed incompressible, isothermal and fully developed.
6. Model Simulation
An implicit steady-state cell based solution procedure was used to solve the governing equation in this study. The SIMPLE algorithm was used for the pressure-velocity coupling and a PRESTO scheme was imposed to result the pressure interpolation. A 2nd order upwind scheme was choose for the solution of the momentum equations and the modified HRIC scheme for the solution of the volume fraction equation. The relaxation factors were typically set to 0.2 and the standard K-Epsilon turbulence model with equilibrium wall functions was used to simulate turbulent flow regime. The y+ values for the wall adjacent cells over the solid geometry were in the range of 80 and 100. That allows being comfortable within the guidelines for the use of the wall function approach. Before running the simulation, it is very important to know that Ansys determines to stop the simulation taking in consideration the iteration number and convergent limit that were specified. It means that once the maximum iteration number is reached or the convergence limit is satisfied the computation is terminated. It was imposed for every calculation a convergence limit of 10E-6 for each monitor and a maximum iteration number of 1495. Figure 4. Convergence plot (stopped at 1495 iteration). Simulation initial convergence was found to be considerably enhanced. It was due to the initialization procedure that it has to be careful attended. It is recommended that the primary phase is set to the lower density fluid, as in this case it was the air. The specific operating density should be set to that of the primary phase and the reference pressure location has to be set to the region of the primary phase. Also it is necessary and helpful to initialize the entire water domain with the correct hydrostatic pressure profile and to initialize both phase domains at the same velocity.
6.1 Model Validation
According to experimental results which were derived from literature experimental tests, total drag coefficient is shown in figure 5. The results of series of simulation with Ansys Fluent were completely compared with the given results of experimental results. The simulation was set up to get the same situation and the same size as the ones which were utilized in theoretical experiences and good agreement could be found. Figure 5a.CFD Pressure Coefficient. Figure 5b.Experimental Pressure Coefficient. It was found that the main problem encountered in this study and independently was the air/water interface became considerably disrupted during the course of the simulation run. In some cases large drops of water were found to be suspended in the air in completely irregular patterns. It was necessary to repatching the appropriate part of the domain with the air phase during the course if the simulation. The values for the pressure coefficient predicted by theoretical methods are smaller than those predicted by the CFD model. It can be conclude that the bad alignment of the struts (attack angle – 10°) in the CFD simulation can explain the relatively large values of the pressure coefficient given by the simulation. The pressure coefficient values of CFD simulation using an angle of attack perfectly aligned are closer than those attended from theoretical experiences. Nevertheless, the CFD values of this condition do not capture the effect of the interaction between water and air. Figure 6a.Experimental results. Figure 6a.Experimental results Olivieri et al.  provide figures of the wave pattern and the wave profile along the ship at Fr=0.41 from the towing tank measurements. In figure 7, the experimental results from Olivieri et al. for the wave pattern on the free surface at Fr=0.41, and at the bottom half is the wave pattern that was found from the CFD code at the same physics work conditions. It is noticed that there is a good agreement with the CFD code results for the general form of the wave pattern from the towing tank test. Figure 7. Experimental and CFD wave pattern at Fr=0.41. Figure 8. Dynamic Pressure Contour. Figure.9 below shows the velocity contour at Fr=0.41. The zones with larger values of velocity (drawn with a red or closer to red colour) have a greater contribution of the frictional and pressure resistance values. It can be seen that the pressure coefficient at the tip edges receives larges values as has previously been attended. Figure 9. Velocity Contour. In the towing tank report from literature  there is a photo of the bow pressure contour at Fr=0.41 The visual comparison between the CFD and the towing tank bow contour shows a similar maximum point of pressure in the geometry as shown in figure 8. The goal of this study was to find a CFD simulation that allows compare theoretical results to these given by the theoretical values. In general the results, at the lower speeds, were not good enough. One reason of this result is that for smaller parameter values at lower speeds more accurate numerical predictions relative to the higher speeds are required to get 5% accuracy compared to the experimental data. Olivieri et al.  provide figures of the wave pattern and the wave profile along the ship at Fr=0.41 from the towing tank measurements. It can be seem in figure.7 the experimental results from Olivieri et al. for the wave pattern on the free surface at Fr=0.41, and at the bottom half is the wave pattern that was found from the CFD code at the same physics work conditions. It is noticed that there is a good agreement with the CFD code results for the general form of the wave pattern from the towing tank test.
The present study presents how possible it is to numerically create a functionally model of a container ship from a simplified geometry. This approximation gives a pretty good estimation of the results attended from theoretical experiences. Some of the scientific papers related to this problem consider the flow as unsteady flow, attached to a cavitation model at high ship velocities. Using that large eddy simulation model as turbulence model could also capture the free surface effect on the hydrofoil as the hydrofoil position during service is around to the surface water. Nevertheless, it was proved that using a K-Epsilon model and considering the flow as two dimensional and fully turbulent flow gives well estimation of the physical working. Numerical approach was used to calculate the free-surface flow around a simplified geometry of a DTMB 5415-1 naval ship model. The simulations were performed using hexahedral mesh after doing some mesh sensibility tests. The results given by the simulation show that this numerical approach is able to accurately simulate the total pressure coefficient, near-field wave shapes, and the pressure and velocity field in the propeller plane. Using Ansys Fluent it can be found that the software is particularly sensitive to the attack angle of the grid respect to the free surface water line in simulation of this kind. This can represent a problem, but I can be solved by re-meshing the geometry for each individual sinkage and trim calculation. The pressure coefficient calculated on the simulation within an attack angle null was found agree with the experimental value to within 8.3%. The simulated velocity and pressure contour shapes along the surface of the ship were found to be in good qualitative agreement to those of experimental profiles. Regarding the possible power saving it should be conclude that for medium speed ranges the results are not changing, while for the angle of attack deviation reach from 0.7 up to 14. Thus, this could be a recommended for better power saving prediction in the speed range considered. Therefore, the predicted power saving are generally lower, but it exhibit a good approximation of real experiences. The above results are encouraging and indicate that Ansys Fluent code is a viable tool and it could be considered for use in more demanding real problems.
8. Recommendation for Future Work
This study was focused on optimizing trim of ship simplified geometry. Next step would be to simulate a more realistic geometry to compare the results obtained in this study. In the direct power saving, that could be considered to predict power levels a wake scale effect and correlation allowance coefficient. References
- M. Reichel, A. Minchev & N.L. Larsen, ‘’Trim Optimisation – Theory and Practice’’, FORCE Technology, Kgs. Lyngby, Denmark, 2014.
- Kodama, Y., Takeshi, H., Hinatsu, M., Hino, T., Uto, S., Hirata, N. and Murashige, S., “Proceedings of the 1994 CFD Workshop”, Ship Research Institute, Japan, 1994.
- Larsson, L., Stern, F. and Bertram, V., “Benchmarking of Computational Fluid Dynamics for Ship Flows: The Gothenburg 2000 Workshop”, Journal of Ship Research, 47, No.1, pp. 63-81, March 2003.
- Hino, T. (ed.), “Proceedings of the CFD Workshop Tokyo 2005”, Tokyo, Japan, 2005.
- Burg, C.O.E, Sreenivas, K., Hyams, D.G. and Mitchell, B., “Unstructured Nonlinear Free Surface Simulations for the Fully-Appended DTMB Model 5415 Series Hull Including Rotating Propulsors”, Proceedings of the 24th Symposium on Naval Hydrodynamics, Fukuoka, Japan, 8-13 July, 2002.
- Wilson, W., Fu, T.C., Fullarton, A. and Gorski, J., “The Measured and Predicted Wave Field of Model 5365: An Evaluation of Current CFD Capability”, presented at the 26th Symposium on Naval Hydrodynamics, Rome, Italy, 17-22 September, 2006.
- ITTC, 1999, “Report of the Resistance Committee,” Proceedings International Towing Tank Conference, Seoul, Korea & Shanghai, China, 5-11 September.
- G2K 2000, http://www.iihr.uiowa.edu/gothenburg2000.
- Stern, F., Longo, J., Penna, R., Olivieri, A., Ratcliffe, T., and Coleman H.: International Collaboration on Benchmark CFD Validation Data for Surface Combatant DTMB Model 5415, Twenty-third Symposium on Naval Hydrodynamics, 401-420, Val de Reuil 2000.
- Olivieri, A., Pistani, F., Avanzini, A., Stern, F., and Penna, R.: Towing tank experiments of resistance,sinkage and trim, boundary layer, wake, and free surface flow around a naval combatant INSEAN 2340 model, IIHR Technical Report No. 421, 2001.
- Heydarian A. and Pengfei Liu, ‘’Ship Trim Optimization for the Reduction of Fuel Consumption’’, 2014.
- D.A. Jones and D.B. Clarke, ‘’ Fluent Code Simulation of Flow around a Naval Hull: the DTMB 5415’’, Maritime Platforms Division Defence Science and Technology Organisation, Victoria, 2010.
- https://www.slideshare.net/AhmedGamal155/hydrofoil-ship-simulation-using-ansys-fluent , (Hydrofoil CFD Workshop)
- INSEAN Research Institute – www.insean.cnr.it
- Lemb Larsen N., Simonsen C.D., Klimt Nielsen C., Råe Holm C. (2012), Understanding the physics of trim, 9th annual Green Ship Technology Conference,Copenhagen, Denmark
- Hansen H., Freund M. (2010), Assistance Tools for Operational Fuel Efficiency, 9th International Conference on Computer and IT Applications in the Maritime Industries, COMPIT 2010, Gubio, Italy
- Hansen H., Hochkirch K. (2013), Lean ECO‐Assistant Production for Trim Optimisation, 11th International Conference on Computer and IT Applications in the Maritime Industries, COMPIT 2013, Cortona, Italy
- Kim, Mingyu and Hizir, Olgun and Turan, Osman and Day, Sandy and Incecik, Atilla (2017) Estimation of added resistance and ship speed loss in a seaway. Ocean Engineering, 141. pp. 465-476. ISSN 0029-8018