New hybrid between NSGA-III with multi-objective particle swarm optimization to multi-objective robust optimization design for Powertrain mount system of electric vehicles

In this study, a new methodology, hybrid NSGA-III with multi-objective particle swarm optimization (HNSGA-III&MOPSO), has been developed to design and achieve cost optimization of Powertrain mount system stiffness parameters. This problem is formalized as a multi-objective optimization problem involving six optimization objectives: mean square acceleration and mean square displacement of the Powertrain mount system. A hybrid HNSGA-III&MOPSO is proposed with the integration of multi-objective particle swarm optimization and a genetic algorithm (NSGA-III). Several benchmark functions are tested, and results reveal that the HNSGA-III&MOPSO is more efficient than the typical multi-objective particle swarm optimization, NSGA-III. Powertrain mount system stiffness parameter optimization with HNSGA-III&MOPSO is simulated, respectively. It proved the potential of the HNSGA-III&MOPSO for Powertrain mount system stiffness parameter optimization problem. The amplitude of the acceleration of the vehicle frame decreased by 22.8%, and the amplitude of the displacement of the vehicle frame reduced by 12.4% compared to the normal design case. The calculation time of the algorithm HNSGA-III&MOPSO is less than the algorithm NSGA-III, that is, 5 and 6 h, respectively, compared to the algorithm multi-objective particle swarm optimization.


Introduction
Due to the different configuration and operating principle of electric motor and internal combustion engine (ICE), the speed and speed of electric motors and ICEs vary greatly, so the dynamic characteristics on electric vehicles will be very different from traditional ICE cars. On the other hand, due to increasing environmental requirements, the big trend is shifting to a dynamical system that uses electricity instead of gasoline as currently. Therefore, our research focuses on the vibration of electric motors in vehicles. The electric vehicle's drive system consists of Powertrain, transmission, and clutch, which is the source of energy for the vehicle and is also one of the major sources of vibration in the car.
There have been some authors studying electric vehicle power systems. [1][2][3] Therefore, to isolate vibrations transmitted from the transmission system to the vehicle body, the Powertrain mount is usually installed between the transmission and the vehicle body. 4 Powertrain mounting system is a system mounted between the frame and the Powertrain. These mounts play an important role in the entire dynamic system of the vehicle and the principle diagram of the full-car dynamic model with the Powertrain mount system is shown in Figure 1. Powertrain mounting system has a suitable stiffness. On the one hand, it will improve the noise, vibration performance, and harshness of the vehicle; on the other hand, it will extend the life of the Powertrain and related parts. 5 If calculated according to the characteristics of controllability, Powertrain racks can be classified into passive Powertrain racks (hydraulic and rubber racks are the most popular), Powertrain racks, and sold Powertrain. Typically, a rubber hanger consists of a metal frame in which the rubber is bonded through adhesives or during vulcanization. With the advantage of low cost and simple structure, rubber suspension is the most widely used engine mount. The important step to designing racks is the parameters that match their stiffness. Suitable rigidity can not only reduce the vibration of the Powertrain to the elastic platform (such as frame and body) but can also reduce the unwanted impact of excitation from the road and impact wheel on the body. One of the most important steps to designing Powertrain racks is to calculate the stiffness parameters of the suspension so that it is best suited to the Powertrain and chassis parameters. With the stiffness parameters of the appropriate Powertrain mount, it can not only reduce the vibration of the Powertrain to the chassis but also reduce the undesirable effects of stimuli from the road and movable impact wheels transfer on the body. By that, the problem of calculating the optimal hardness parameters for the Powertrain mounting system is an important task in designing the vehicle's dynamic system. This is a multi-object concurrency optimization problem.
Recently, a number of researchers have studied in the field of multi-objective optimization. They introduced various methods, among them, presented in Konak et al. 6 in reviews and guidelines. NSGA-II algorithm is published by Deb et al. 7 and so far there are a number of variations and applications of NSGA II algorithm developed by Chang and Chang, 8 Ishibuchi et al., 9 and Malekmohammadi et al. 10 Deb and Jain 11 have published and applied the MONGA-II method for a number of multi-objective testing problems. NSGA-III algorithms have been studied to face multiple goals at once (more than two). This is the Algorithm published by Deb and Jain 12 in 2014, in which they changed some selection mechanisms. They came up with a multi-objective evolution algorithm based on reference points based on the NSGA-II algorithm. They mainly emphasize that population members are not popular, but close to the combination of a set of reference points provided. The NSGA-III algorithm is proposed to apply to a number of multiobjective testing problems with 3 to 15 goals.
In addition, there are some researchers studying the optimal problem of many objects. They have studied and developed multi-objective particle swarm optimization (MOPSO) algorithm. MOPSO algorithm is one of the most popular multi-objective optimization algorithms conceptually; it is similar to particle swarm optimization (PSO). Coello et al. 13 and his colleagues applied MOPSO algorithm to handle multiple objectives. In recent studies of MOPSO algorithm, 13-15 they have shown additional conditions such as multiple estimates used to achieve better exploration characteristics. Baltar and Fontane 16,17 improved the MOPSO algorithm to minimize deviation from outflow water quality. They also published an application of an evolutionary optimization algorithm for multi-objective analysis for reservoir operations and planning. Reddy and Kumar 18,19 published and applied the Elitist-Mutated operator with MOPSO (EM-MOPSO) to show the reduction of total squared deviations for irrigation, maximizing the yield of aquatic electricity and the degree of satisfaction of downstream water quality requirements. In addition, they used the Elitist-Mutated MOPSO algorithm (EM-MOPSO) to maximize hydropower production and minimize the total number of squares to release annual irrigation from demand. Wang et al. 20 have applied modified MOPSO to minimize the highest water level, release peak flow, water level difference after flood season, and flood control level. In this case, the concept of Pareto dominance for selecting leaders from a non-dominated external archive has been utilized by MOPSO algorithm where the leaders of swarms that guide the particles to the Recently, there are some new optimization algorithms proposed. [21][22][23][24][25][26][27] However, each algorithm has advantages and disadvantages, precisely, because no algorithm can solve all optimization problems correctly. Therefore, new hybrid algorithms should be proposed to be able to solve new problems that have not been resolved before and/or have better accuracy than existing techniques.
On the other hand, there are some hybrid methods of optimization algorithms that have been recently developed: Jeong et al. 28 published the development and investigation of the GA/PSO-hybrid algorithm effectively for optimizing the design in the real world. Premalatha and Natarajan 29 published hybrid PSO and GA for Global Maximization. A multi-objective particle optimization method based on extreme optimization with variable and inertial inertia mutations (HM-TVWF-MOEPSO) has been proposed to solve some of the problems in optimization, multi-purpose particle chemistry, and improved algorithm performance. 30 A new hybrid heuristic algorithm is published in the current work for multi-objective optimization issues. The hybrid algorithm has proposed a method to combine the simple algorithm Nelder-Mead with the nondominant genetic algorithm II (NSGA II) to find the best global point. The performance of this new algorithm has been presented through a number of complex benchmark functions. 31 A pre-selected pre-creation method has been published to address multivariate technical optimization problems. 32 This method can set the number of Pareto solutions and optimize multiple times until satisfactory results are obtained. This is an effective algorithm that consists of independent parallel genetic algorithms by dividing the entire population into multiple populations, 33 in which each population group will be assigned to different weights to search for optimal solutions in different directions. Therefore, most published hybrid algorithms have many advantages. This breeding has overcome the limitations of each optimization algorithm. This proves that this is one of the methods that should be studied in multiobjective optimization.
In this article, a new hybrid optimization algorithm is proposed in this work for multi-objective problems. This is the hybrid between the MOPSO algorithm and a multi-objective genetic algorithm (NSGA-III) to find the best of the Pareto optimal front. HNSGA-III&MOPSO is proposed to outperform MOPSO and NSGA-III because it uses a combination of search operators of both algorithms to create a new population. This makes the search process more diverse, wider search space. New hybrid algorithms show better performance than other algorithms. This is demonstrated through a number of complex benchmarking functions and Powertrain mount system stiffness parameter optimization problem with six-objective optimization in a three-dimensional (3D) model. The amplitude of the acceleration of the vehicle frame decreased by 22.8%, and the amplitude of the displacement of the vehicle frame reduced by 12.4% compared to the normal design case. The calculation time of the algorithm HNSGA-III&MOPSO is less than the algorithm NSGA-III, that is, 5 and 6 h, respectively, compared to the algorithm MOPSO.
The organization of this article is as follows. Section ''Structure'' describes the proposed hybrid HNSGA-III&MOPSO method and computational experimentation with several benchmark functions. Section ''Vibration characteristic of the Powertrain mount system'' describes the vibration characteristic of the Powertrain mount system and simulation results of application HNSGA-III&MOPSO method to optimization of the Powertrain mount system stiffness

Structure
Genetic algorithm NSGA-III This algorithm was published by Deb and Jain 12 in 2014 with a number of change mechanisms selected. NSGA-III algorithm is based on the steps described in Figure 2.

PSO (MOPSO)
Kennedy et al. 34 published an algorithm based on the basis of PSO algorithm for optimization. They improved the PSO algorithm to find the Pareto optimal front. Therefore, the improved PSO algorithm is suitable to optimize many goals with high convergence speed, allowing each individual to benefit from the experience. A diagram of MOPSO is shown in Figure 3.

Hybrid NSGA-III and MOPSO (HNSGA-III&MOPSO)
Each evolutionary algorithm has different strengths and characteristics. Therefore, it is only natural to think of integrating different algorithms to handle a complex problem. In the field of research, the evolution algorithm integrates two or more optimization algorithms into a single frame. The results show that hybrid algorithms have higher efficiency because they can exchange characteristics to improve the disadvantages and enhance their advantages. Parallel hybridization can improve exploration and exploitation which can yield higher performance and more favorable conditions than any single algorithm. These populationbased approaches use different techniques to explore the search space and, when they are combined, will improve the trade-off between exploration and exploitation tasks to converge around. The best solution was possible.
HNSGA-III&MOPSO hybrid approach. HNSGA-III&MOPSO is implemented in parallel breeding; that is, the initial population will be generated in both NSGA-III and MOPSO. After that, two separate populations will be mixed together. The new population after combining will be both algorithms used as their own population to perform fitness function calculations that evaluate the evolution of each algorithm. By the next generation, the new population created by the two algorithms is mixed together to form a common population. The process repeats until the end of evolution condition is completed. The process of parallel operation is to create an extremely diverse and widespread population. This makes the algorithm have a wide search strategy across the regions, besides making the process more convergent. Therefore, the analysis time is reduced and the results are more accurate. The pseudo code of the proposed algorithm: Parameter initialization for NSGA-III and MOPSO algorithms. MOPSO algorithms: Swarm population initialization. NSGA-III algorithms: GA population initialization. While travel not completed.
Combination the two populations. MOPSO algorithm. While sub-travel not completed. Determined fitness function. Set P best and G best. Update particle velocity v i k + 1 and position x i k . Proceed non-dominated sorting and crowding distance.
While evolution not completed. Choose two parents P1 and P2 using the tournament method.

(continued)
A flow chart of HNSGA-III&MOPSO is shown in Figure 4.

Computational experimentation with several benchmark functions
Numerical results. In this section, the performance of HNSGA-III& MOPSO is evaluated using five benchmarks that are published in CEC 2009, [35][36][37][38] listed in Table 1. The results are compared to algorithms of NSGA-III and MOPSO. For the performance metric, 22 Inverted Generational Distance (IGD), 39 Spacing (SP), and Maximum Spread (MS) 40 criteria are employed to measure convergence, quantity, and coverage, respectively. The mathematical formulation of IGD is as follows where N is the number of true Pareto optimal solutions and d i indicates the Euclidean distance between the ith true Pareto optimal solution and the closet obtained Pareto optimal solutions in the reference set. It should be noted that IGD = 0 indicates that all members of the non-dominated solutions are in the true Pareto Front Proceed the crossover between P1 and P2 with a probability Pc.
Recognize the non-dominated population sorting. Normalize the population members. Associate the population member with the reference Points.
Apply the niche preservation (counter).
Keep the appropriate solutions obtained for the next generation.
UF8: f 1 = cos (0:5x and J 3 are the same as those of UF10 UF10: where d is the average of all d i , N is the number of Pareto optimal solutions obtained, and where d is a function to calculate the Euclidean distance, b i is the minimum in the ith objective, and M is the number of objectives. In addition to using performance metrics, the best Pareto optimization set that HNSGA-III&MOPSO obtained on both parameter space and search space is shown in Figures 5 and 6    in Tables 5-7. IGD shows that the proposed hybrid algorithm (HNSGA-III&MOPSO) can provide the best results on all statistics for issues that test two goals. IGD is a performance indicator that shows the accuracy and convergence of the algorithm. Therefore, it can be said that the proposed HNSGA-III&MOPSO algorithm can provide outstanding convergence on benchmarking two or three optimal goals. Pareto optimal solution results of HNSGA-III&MOPSO on each benchmark are also described in Figures 5 and 6.
The resulting Pareto front is shown in Figures 6  and 7.
The numerical results prove that HNSGA-III&MOPSO having good performance for optimal objects is two objects; it relates to the convergence and scope of the search. However, HNSGA-III&MOPSO having good performance for optimal objects is three objects; the proposed algorithm shows high convergence and better coverage of many MOPSO and NSGA-III algorithms. From here, we can say that the main advantages of the algorithm HNSGA-III&MOPSO proposed compared to NSGA-III and MOPSO are the convergence characteristics and the ability to search more broadly. In addition, the results  of HNSGA-III&MOPSO are proposed in most cases better than MOPSO and NSGA-III. Therefore, the results show that HNSGA-III&MOPSO is proposed to outperform MOPSO and NSGA-III because it uses a combination of search operators of both algorithms to create a new population. In addition, the MOPSO algorithm updates the gBest in each iteration. Therefore, all particles are attracted by the same or a similar gBest (i) group in each iteration, while individuals of HNSGA-III&MOPSO are updated for each generation, thus supporting assist search agents to explore a wider search space. This proves that the hybrid method has actually been successful. It gives more accurate results and less search time than the other two algorithms.
Vibration characteristic of the Powertrain mount system Using Newton's law, the mathematical model of Figure 1 can be written as follows where the symbols are shown in Table 8.

Multi-objective optimization functions
There are many indicators to evaluate the vibration of the Powertrain. In particular, mean square acceleration oscillates at the front and rear of the Powertrain mount, the mean square displacement difference between the Powertrain and vehicle chassis at the front and rear Powertrain mount. These are two important parameters that determine the decisive influence of unit Powertrain vibration on chassis. In order to optimally reduce the vibration of the unit Powertrain, we need to simultaneously optimize the parameters of mean square acceleration and mean square displacement at the front and rear, right and left of the Powertrain mounts. The average square value of the vibration acceleration of any points can be determined by the following formula where v is the frequency, jW z (jv)j 2 is the squared modulus of amplitude and phase characteristics, and S q (v) is the spectral density of exposure.  The matrix of inertial coefficients of car parts C The matrix of coefficients of stiffnesses and torsional rigidity K The matrix of damping coefficients Q(t) Column vector of the perturbing forces and moments q 1 , q 2 Universal road surface amplitude at front and rear wheels q 2 (t) = q 1 (t + t) with t: time interval, va: vehicle speed The modules of the amplitude and phase characteristics of the vibratory displacement of the Powertrain and the car body in the place of the front mount are where W 0 (jv) is the module amplitude-phase characteristics in the center of the car body, W f 0 (jv) is the module amplitude-phase characteristics of the longitudinalangular body of the car, W O 1 k (jv) is the module of the amplitude and phase characteristics of the body in place of the front mount of the Powertrain, W ca (jv) is the module amplitude-phase characteristics in the center of the Powertrain of the car, W f ca (jv) is the module amplitude-phase characteristics of the longitudinalangular Powertrain of the car, and W O 1 ca (jv) is the modulus of amplitude-phase characteristics in place of the front mount of the Powertrain.
The modules of the amplitude-phase characteristics of the vibration displacement of the Powertrain and the car body in the place of the rear mount are where W O 2 k (jv) is the module of the amplitude-phase characteristics of the body in the place of the rear mount of the Powertrain and W O 2 ca (jv) is the module of amplitude-phase characteristics in the place of the rear mount of the Powertrain.
The difference of the module frequency response of the vibration of the Powertrain and the car body in place of the front mount of the Powertrain is The difference of the frequency response module of the vibration displacement of the Powertrain and the car body at the rear mount is Simulated input parameters Road surface profiles. When the vehicle moves, there are many factors that cause the vibration:: the internal force in the car; external forces that appear in the process of using acceleration, braking, and revolving; exterior conditions such as wind and storm; and boring face street. Among the factors on the bumpy side of the road is the oscillation cause of the vehicle. To simulate the most general calculation, we use the road surface profile as a random function as in Figure 7 and simulated parameters as shown in Table 9.
Simulation results of application HNSGA-III&MOPSO method to optimization of the Powertrain mount system stiffness parameter   It is well known that the results of multi-objective optimization would be a set of non-dominated optimized points called Pareto set. These points offer a wide range of parameters to the designer to choose the optimum point depending on his designing conditions. There are always conflicting objective functions in vehicle designing where improvement in one function may have an unfavorable influence on other functions. In this article, multi-objective optimization for all sixobjective functions is done simultaneously. Application of HNSGA-III&MOPSO optimization algorithm: we obtain results as shown in In which f1 is the Pareto front of MSD at the front left Powertrain mount, f2 is the Pareto front of MSA at the front left Powertrain mount, f3 is the Pareto front of MSD at the front right Powertrain mount, f4 is the Pareto front of MSA at the front right Powertrain e mount, f5 is the Pareto front of MSD at the rear Powertrain mount, and f6 is the Pareto front of MSA at the rear Powertrain mount.
HNSGA-III&MOPSO has been applied in the problem of Powertrain mount system stiffness parameters optimization. Simulation results comparing one of the results in the set of the Pareto front from the HNSGA-III&MOPSO algorithm with different stiffness values (K1.1, K1.2, K1.3) are shown in Figures 11 and 12. Figure 11 shows the acceleration of the vehicle frame corresponding to the different stiffness values. Symbol A corresponds to the optimal stiffness value in a set of the Pareto front. Symbol B corresponds to K1.1 = 290,000, K1.2 = 350,000, K1.3 = 550,000. Symbol C corresponds to K1.1 = 790,000, K1.2 = 750,000, K1.3 = 350,000. From the graph, we see that the smallest acceleration is at the optimal stiffness value.
Similarly, Figure 12 shows the displacement of the vehicle frame corresponding to the different stiffness values. Symbol A1 corresponds to the optimal stiffness value in a set of the Pareto front. Symbol A2 corresponds to K1.1 = 290,000, K1.2 = 350,000, K1.3 = 750,000. Symbol A3 corresponds to K1.1 = 790,000, K1.2 = 750,000, K1.3 = 650,000. From the graph, we see that the smallest displacement is at the optimal stiffness value in a set of the Pareto front.