A sensor layout method for fault diagnosis of piston and connecting rod systems

Abstract

This invention discloses a sensor layout method for fault diagnosis of piston-connecting rod systems in diesel engines. Data is collected under each pre-designed sensor layout scheme. A data preprocessing module obtains the sensor feature vectors, which are then used as input to an LSTM neural network. The actual fault and normal operating condition numbers are used as outputs to train a fault diagnosis model. A test dataset is used as input to the fault diagnosis model, and the predicted operating condition number is output. The fault diagnosis accuracy is calculated. The fault diagnosis accuracy and the feature vectors corresponding to each scheme are used as input to a binary Skyhawk algorithm sensor network optimization module, with the binary code of the optimal sensor layout as the output. The method uses the number of sensors, information entropy, and fault diagnosis rate as evaluation indicators to remove redundant information, reduce the number of sensors, and ensure accurate fault diagnosis.

Description

Technical Field

[0001] This invention relates to the field of sensor network layout optimization and fault diagnosis, and in particular to a sensor network system and method for fault diagnosis of piston connecting rod system in diesel engines. Background Technology

[0002] The piston-connecting rod system, a core component for the normal and efficient operation of a diesel engine, plays a crucial role in converting the linear reciprocating motion of the piston into the rotary motion of the crankshaft. Currently, the operating status information of the piston-connecting rod system mainly relies on its sensor network. However, the number and location of sensors have a significant impact on the performance and reliability of the monitoring system. An inappropriately configured sensor network can lead to increased costs and reduced accuracy in troubleshooting the monitoring system.

[0003] The sensor network optimization layout disclosed in Chinese Patent Application Publication No. CN116341373A, entitled "A Sensor Network Optimization Layout Method," obtains the overall modal information of the structure by performing modal simulation analysis on structural deformation, initializes the number and layout positions of sensors, and optimizes the layout by utilizing the excellent local search capabilities of the particle swarm optimization algorithm and the excellent global search capabilities of the fruit fly algorithm. The sensor network layout disclosed in Chinese Patent Application Publication No. CN109751113A, entitled "Sensor Network Layout Optimization Method for Microseismic Monitoring in Mines and Its Application," constructs an optimization function for the sensor network layout of an unknown wave velocity system using the CRLB principle of optimal parameter estimation, and solves the optimization function using an improved genetic algorithm with encoding operators, selection operators, crossover operators, mutation operators, elite retention strategies, and convergence criteria to obtain the optimal sensor network layout scheme. However, when a diesel engine is operating, its piston-connecting rod system is in a non-open state, making structural deformation simulation impossible. Therefore, these methods are not suitable for piston-connecting rod systems that undergo linear reciprocating motion. Summary of the Invention

[0004] The purpose of this invention is to solve the problem that existing sensor network layouts are not suitable for piston-connecting rod systems that perform linear reciprocating motion, and to propose a sensor network system and method for fault diagnosis of piston-connecting rod systems, so as to achieve accurate diagnosis of piston-connecting rod faults.

[0005] The present invention provides a fault diagnosis method for a sensor network system for piston and connecting rod system fault diagnosis, which adopts the following technical solution:

[0006] Step 1): Pre-layout based on experience 2 n -1 sensor layouts, where n is the number of units under test. Data for each sensor is collected under each pre-layout scheme and divided into training and test datasets. The feature vectors of the sensors are obtained through the data preprocessing module.

[0007] Step 2): Use the feature vector as the input to the LSTM neural network, and the actual fault and normal operating condition numbers of the piston connecting rod system as the output to train the fault diagnosis model.

[0008] Step 3): Using the aforementioned test dataset as input to the fault diagnosis model, the model outputs a predicted operating condition number, and the result is calculated as 2. n - Fault diagnosis accuracy under one sensing layout;

[0009] Step 4): Use the fault diagnosis accuracy and the feature vector corresponding to each scheme as input to the binary Eagle algorithm sensor network optimization module, and use the binary code of the optimal sensor layout as output.

[0010] Furthermore, the specific method for step 4) is as follows:

[0011] (1) Set the parameters of the binary eagle algorithm, randomly allocate the population size, and generate a set of random n-dimensional binary codes for each individual;

[0012] (2) Calculate the information entropy, the proportion of sensors deployed in binary codes, and the fitness value of the feature vector corresponding to each randomly generated binary code;

[0013] (3) Sort all individuals in descending order of fitness value and select the individual with the highest fitness value as the current best position;

[0014] (4) Update the population position. After updating, normalize the position of each individual in the population. Perform binary encoding mapping on the normalized position and randomly generate a set of n-dimensional [0,1] vectors. In the normalized individual position, if the value of a certain dimension of the normalized individual position is less than the corresponding value of the corresponding dimension of the vector, then the value of that dimension is set and no sensor is installed; otherwise, the sensor is installed.

[0015] (5) Recalculate the updated fitness value for the individual positions in the mapped population, compare the updated fitness value of the population with the current best fitness value, and if the fitness value in the updated generation is better than the current best fitness value, then take the updated best population individual position and the corresponding fitness value as the current best fitness value and the best population individual position.

[0016] (6) Determine whether the maximum number of iterations has been reached. After the iteration is completed, the position of the Sky Eagle population is the binary code of the optimal sensor network.

[0017] Furthermore, when updating the population position, the maximum number of iterations is H. If the current iteration number h ≤ 2H / 3, and the random number rand in the range [0,1] ≤ 0.5, the updated individual position is:

[0018]

[0019] If rand > 0.5, the updated individual position is:

[0020] X2(h+1)=X best (h)×Levy(n)+X R (h) + (po) × rand,

[0021] X best (h) represents the optimal solution up to the h-th iteration, X M (h) represents the average position of the population at the h-th iteration, Levy(Dim) is the Levy flight distribution function, and X R (h) is a random solution taken in the range [1, Num] for the h-th iteration, where Num is the population size, and p and o represent the search spiral shape.

[0022] If the current iteration number h > 2H / 3, and the random number rand in the range [0,1] > 0.5, the updated individual position is:

[0023] X3(h+1)=(X best (h)-X M (h))×a-rand+((UB-LB)×rand+LB)×a;

[0024] If rand ≤ 0.5, the updated individual position is:

[0025] X4(h+1)=QF×X best (h)-(G1×X(h)×rand)-G2×Levy(n)+rand×G1,

[0026] α is the mining adjustment parameter set to 0.1, QF represents the quality function used to balance the search strategy, G1 represents the various actions a predator takes to track its prey during the hunting process, and G2 represents the flight slope used to follow the prey.

[0027] The advantages of this invention using the above technical solution are:

[0028] 1) The present invention provides a sensor network system for fault diagnosis of piston and connecting rod systems, which integrates modules such as data preprocessing and binary Skyhawk algorithm sensor network layout optimization, thereby reducing the complexity of the central processing unit and improving the reliability and economy of fault diagnosis of piston and connecting rod systems.

[0029] 2) The Tianying algorithm and sensor network layout optimization modules adopted solve the problems of long modeling time and complex modeling process in traditional mechanism analysis-based sensor placement for multi-coupled and nonlinear systems.

[0030] 3) The present invention provides a sensor layout method for fault diagnosis of piston connecting rod systems. The method uses the number of sensors, information entropy, and fault diagnosis rate as evaluation indicators. While ensuring the fault diagnosis rate, it removes redundant information, reduces the number of sensors, lowers costs, improves monitoring accuracy and computational efficiency, and ensures accurate fault diagnosis. Attached Figure Description

[0031] Figure 1 This is a schematic diagram of a piston-connecting rod system;

[0032] Figure 2 for Figure 1 A schematic diagram of the structural stress analysis of the piston-connecting rod system in the diagram;

[0033] Figure 3 This is a structural block diagram of a sensor layout method for fault diagnosis of piston-connecting rod systems according to the present invention.

[0034] Figure 4 This is a binary encoding diagram of a sensor network;

[0035] Figure 5 for Figure 3 Flowchart of the data preprocessing module execution;

[0036] Figure 6 for Figure 3 The flowchart of the execution of the sensor network optimization module of the binary Skyhawk algorithm. Detailed Implementation

[0037] like Figure 1 The piston-connecting rod system shown mainly includes piston 1, connecting rod 2, bearing 3, and crankshaft 4. It is a key power transmission component in a diesel engine, and its operation includes intake stroke, compression stroke, power stroke, and exhaust stroke. Connecting rod 2, acting as the connector between piston 1 and crankshaft 4, converts the linear reciprocating motion of piston 1 into the continuous rotational motion of crankshaft 4. Bearing 3 is located at the pivot point of crankshaft 4, ensuring smooth operation of crankshaft 4.

[0038] like Figure 2 The diagram shows the forces acting on the piston-connecting rod system. Point O is the center of rotation of crankshaft 4, point B is the center of the piston pin, and the crank OA of crankshaft 4 rotates. When the piston-connecting rod assembly reciprocates within the cylinder, the total force on the piston pin is F. Let D be the cylinder liner diameter, m. j Let R be the reciprocating inertial mass of the piston-connecting rod assembly, ω be the crank radius OA, ω be the crank angular velocity, θ be the crank angle, λ be the ratio of crank radius to connecting rod length, and ν be the inertial mass of the piston-connecting rod assembly. h (t) represents the lateral velocity of the piston, a hLet δ be the lateral acceleration of the piston, and δ be the distance between the piston and the cylinder liner. Then, the total force F at the piston pin can be decomposed into the following horizontal force F along the horizontal direction. h and the force F along the direction of the link c F h This refers to the thrust of the piston on the cylinder liner:

[0039]

[0040] The lateral impact momentum M of the piston connecting rod assembly on the cylinder liner j for:

[0041]

[0042] From the above formula, we can see that the impact momentum M of the piston connecting rod assembly on the cylinder liner is... j The vibration is related to the engine speed and time t. When the diesel engine is running, when piston 1 reaches top dead center, it is affected by inertial force, which changes its motion tendency. At this time, due to the change in the direction of piston 1's motion and the force of the piston-connecting rod assembly, the left side of piston 1 is prone to impact with the cylinder wall. Similarly, at bottom dead center, the right side of piston 1 is prone to impact with the cylinder wall. Therefore, based on the analysis of the minimum vibration transmission path, for piston ring and cylinder liner wear faults in a single-cylinder diesel engine, at least one vibration sensor needs to be installed on the cylinder head for monitoring.

[0043] When piston 1 malfunctions, the thrust of connecting rod 2 on crankshaft 4 is F. c Its direction is through the center of connecting rod 2 and acts at the center A of the journal, which can be decomposed into two mutually perpendicular forces F. n and F t This wear causes bearing 3 to begin to wear. After bearing 3 has undergone a certain wear stage, the effective average oil film thickness of the lubricating oil decreases, accompanied by increased lubricating oil leakage and increased heat dissipation. Furthermore, severe wear may cause the main journal to directly contact the bearing bush in poorly lubricated areas, generating an impact effect. Therefore, for a single-cylinder diesel engine, in addition to the above-mentioned vibration sensor arrangement, two more sensors are needed to monitor the vibration and temperature changes of bearing 3. Thus, for an N-cylinder diesel engine, at least 3N test units are required.

[0044] like Figure 3 As shown, for a diesel engine with n test units, each test unit can be connected to a corresponding sensor. Setting a sensor for each test unit would lead to data redundancy and increased system complexity, thus affecting the rapid and accurate acquisition of key information. Therefore, this invention proposes a sensor layout method for fault diagnosis of piston-connecting rod systems to achieve more efficient and accurate monitoring.

[0045] like Figure 4As shown, the sensor placement problem can be viewed as a combinatorial binary optimization problem. Whether or not a sensor is installed on each measured unit can be determined by a binary character vector Vectors = (S1, S2, ..., S...). n ) represents the elements S1, S2, ..., S of the vector. n S represents the unit under test. i =0 indicates that the i-th measured unit does not have a sensor installed; conversely, S i =1 indicates that the sensor is installed in the i-th measured unit, and 1≤i≤n.

[0046] Specifically, taking an N-cylinder diesel engine as an example, based on Figure 1-2 Based on the above empirical analysis, for monitoring a piston-connecting rod system, there are at least 3N units to be measured, where 3N ≤ n, 2 n -1 sensor layout scheme. Based on the above experience analysis, each sensor is pre-laid out. Under each pre-laid-out scheme, data is collected for each sensor. After the data preprocessing module calculates the data, the binary Eagle algorithm sensor network optimization module is used to find the binary code of the optimal sensor layout. This code is then transmitted to the central processing unit via the bus communication module. The central processing unit performs reverse parsing and determines the optimal sensor layout, such as... Figure 3 As shown.

[0047] like Figure 5 The diagram shows the data preprocessing execution process, with 2... n In one sensing layout scheme, data collected by each sensor is used as input, and fault diagnosis accuracy and feature vector are used as output. In 2... n Under one sensor layout scheme, the operating conditions of each sensor under each layout scheme were collected, resulting in s fault and normal operating conditions, corresponding to actual fault and normal operating condition numbers 1, 2, ..., s. Each operating condition had 200 samples, resulting in 200×s m-dimensional samples, where m is the number of sensors in the current layout. 80% of the data from these 200×s m-dimensional samples were used as the training dataset, and 20% as the test dataset. Then, the data collected by each sensor in both the training and test datasets were decomposed using wavelet packet transform technology at three levels, generating 8 frequency bands. The energy entropy of the signal in each frequency band after decomposition was calculated as the feature vector of that sensor, forming an 8×m-dimensional feature vector Z(z1, z2, ..., z...). 8×m The feature vector Z(z1,z2,...,z) is extracted. 8×m This eigenvector Z(z1,z2,...,z) 8×m ) serves as the input to the LSTM neural network and binary Skyhawk algorithm sensor network optimization module.

[0048] In the LSTM neural network, the initial learning rate is set to 0.01. After 60 training iterations, the initial learning rate is adjusted with a learning rate adjustment factor of 0.2 and a regularization parameter of 0.001. The maximum number of training iterations is 100. Finally, the 8×m-dimensional feature vector of the training set is used as the input to the LSTM neural network, while the actual fault and normal operating condition numbers 1, 2, ... s of the piston connecting rod system are used as the output. After sufficient training, a fault diagnosis model is obtained. Then, a pre-divided test dataset is used as input, and the predicted operating condition number is output through the fault diagnosis model. Finally, the 2 is calculated using formula (5). n -1 Fault diagnosis accuracy F(2) under a sensor layout n -1):

[0049]

[0050] Where d represents the number of samples where the output predicted operating condition number differs from the actual fault and normal operating condition numbers 1, 2, ... s when the test dataset is validated.

[0051] The fault diagnosis accuracy F(2) n -1) Input into the Tianying algorithm sensor network optimization module.

[0052] like Figure 6 The diagram shows the execution flowchart of the binary Eagle algorithm sensor network optimization module. n -1 The sensor layout scheme is built into the binary Eagle algorithm sensor network optimization module. The binary Eagle algorithm sensor network optimization module achieves a fault diagnosis accuracy of F(2). n -1) and the feature vector Z(z1,z2,...,z) corresponding to each scheme 8×m The input is the binary code of the optimal sensor layout, and the output is the binary code of the optimal sensor layout. The specific process is as follows:

[0053] (1) Set the parameters of the binary eagle algorithm: population size Num, maximum number of iterations H, population dimension equal to the number of units under test, search upper limit UB, search lower limit LB, and randomly initialize the position of the eagle individuals.

[0054] (2) Randomly allocate the population size and generate a random n-dimensional binary code for each individual, which represents the position of each individual in the population. The position of each individual in the population is an n-dimensional vector, which is then an n-dimensional binary code.

[0055] (3) Calculate the information entropy G(2) of the feature vector corresponding to each randomly generated binary code using formula (6). n -1):

[0056]

[0057] Where, for the feature vector Z(z1,z2,...,z) 8×m ), P(z e ) is state z e The probability distribution, probability distribution P(z) e )Calculate using formula (7):

[0058]

[0059] z j These are the eigenvalues ​​corresponding to different dimensions of the eigenvector, j = 1, 2, ..., 8 × m.

[0060] (4) Calculate the proportion of sensors with randomly generated binary codes using formula (8):

[0061] K(2 n -1)=m / n (8)

[0062] (5) The proportion of sensor arrangement based on binary encoding K(2) n -1) Fault diagnosis accuracy F(2) n -1) and information entropy G(2) n -1), calculate the fitness value fit(2) using formula (9). n -1), to assess the performance of each individual in the current population:

[0063] fit(2 n -1)=2-F(2 n -1)-G(2 n -1)+K(2 n -1) (9)

[0064] (6) All individuals are fitted according to their fitness value (2). n -1) Sort from largest to smallest, and select the individual with the highest fitness value as the current optimal position X. fit The corresponding fitness value is the current best fitness value, fit.

[0065] (7) Update the population position. If the current iteration number h ≤ 2H / 3, and the random number rand in the range [0,1] ≤ 0.5, enter the expanded exploration phase X1. This is equivalent to exploring the search space from a global perspective to find possible solution regions. In this way, the algorithm can quickly identify the region where the potential optimal solution is located in a large range. The specific update formula (10) is as follows:

[0066]

[0067] In the formula, X1(h+1) represents the individual position updated by the first search method, i.e., the expanded exploration phase X1, in the (h+1)th iteration of the Skyhawk algorithm. best (h) represents the optimal solution up to the h-th iteration, X M (h) represents the average position of the population at the h-th iteration, H represents the maximum number of iterations, and rand is a random number in the range [0,1].

[0068] (8) If the current iteration number h ≤ 2H / 3 and rand > 0.5, then enter the narrowing exploration phase X2. This step represents that after finding a potential solution, the algorithm will focus more on local search in this region to approximate the optimal solution. The specific update formula (11) is as follows:

[0069] X2(h+1)=X best (h)×Levy(n)+X R (h)+(po)×rand (11)

[0070] In the formula, X2(h+1) represents the individual position updated by the second search method, i.e., the narrowing exploration phase X2, in the (h+1)th iteration of the Skyhawk algorithm; Levy(Dim) is the Levy flight distribution function; n is the population dimension; X R (h) is a random solution taken in the range [1, Num] for the h-th iteration; p and o represent the shape of the search spiral.

[0071] (9) When the number of iterations h>2H / 3 and rand>0.5, the algorithm enters the expansion development stage (X3). This means that during the search process, the algorithm will take a more meticulous and slow approach to gradually approach the optimal solution. The specific update process is shown in formula (12).

[0072] X3(h+1)=(X best (h)-X M (h))×a-rand+((UB-LB)×rand+LB)×a (12)

[0073] Where X3(h+1) represents the individual position updated by the third search method (X3) in the h+1th iteration of the Skyhawk algorithm, and α is the mining adjustment parameter of 0.1;

[0074] (10) When the current iteration number h > 2H / 3 and rand ≤ 0.5, enter the narrowing development stage X4. The algorithm performs final fine-tuning and confirmation to ensure that the true optimal solution is found. The specific update formula (13) is as follows:

[0075] X4(h+1)=QF×X best(h)-(G1×X(h)×rand)-G2×Levy(n)+rand×G1 (13)

[0076] Where X4(h+1) represents the individual position updated by the fourth search method, namely the narrowing development stage X4, in the h+1th iteration of the Skyhawk algorithm; QF represents the quality function used to balance the search strategy; G1 represents the various actions used by predators to track prey during the hunting process; and G2 represents the flight slope for following the prey.

[0077] (11) After the population update is completed, the position of each individual in the population is normalized using formula (14):

[0078]

[0079] In the formula, X b (h+1) represents the individual position updated by the b-th search method in the (h+1)-th iteration of the Skyhawk algorithm (where b = 1, 2, 3, 4). The normalized position of the updated individual position is T(X). b (h+1)), at which point the position of each individual is an n-dimensional vector of (0-1).

[0080] (12) Using formula (15), the normalized position T(X) is obtained. b (h+1)) is used for binary encoding mapping, and a set of n-dimensional [0,1] vectors r is randomly generated. At the normalized individual position T(X) b In (h+1), each individual's position is compared with the random vector r. If the value of a certain dimension of the normalized individual's position is less than the corresponding value of the corresponding dimension of vector r, then the value of that dimension is set to 0, and no sensor is installed; otherwise, it is set to 1, and a sensor is installed.

[0081]

[0082] In the formula, X ij This represents the location of an individual in the population after mapping.

[0083] (13) Recalculate the updated fitness values ​​for the individual positions in the mapped population using steps (3-5). Since the individual positions are now mapped into a set of binary codes, sort the fitness values ​​and find the updated optimal population individual positions and their corresponding fitness values. Compare the updated population fitness values ​​with the current optimal fitness value `fit`. If the fitness value in the updated generation is better than the current optimal fitness value `fit`, then use the updated optimal population individual positions and their corresponding fitness values ​​as the current optimal fitness value and optimal population individual positions.

[0084] (14) Determine whether the maximum number of iterations H has been reached. If yes, stop the iteration; otherwise, return to steps (3)-(13).

[0085] (15) After the iteration is completed, the position of the Sky Eagle population is the binary code of the optimal sensor network.

[0086] The optimal sensor layout code is found, and this code information is transmitted to the central processing unit via the bus communication module. The central processing unit performs reverse analysis to determine the sensor layout, thereby improving monitoring accuracy.

Claims

1. A sensing layout method for fault diagnosis of piston-connecting rod systems, characterized in that... Includes the following steps: Step 1): Pre-layout 2 n -1 sensor layouts, where n is the number of units under test. Data for each sensor is collected under each pre-layout scheme and divided into training and test datasets. The feature vectors of the sensors are obtained through the data preprocessing module. Step 2): Use the feature vector as the input to the LSTM neural network, and the actual fault and normal operating condition numbers of the piston connecting rod system as the output to train the fault diagnosis model. Step 3): Using the aforementioned test dataset as input to the fault diagnosis model, the model outputs a predicted operating condition number, and the result is calculated as 2. n - Fault diagnosis accuracy under one sensing layout; Step 4): Use the fault diagnosis accuracy and the feature vector corresponding to each scheme as input to the binary Eagle algorithm sensor network optimization module, and use the binary code of the optimal sensor layout as output.

2. The sensing layout method for fault diagnosis of piston-connecting rod systems according to claim 1, characterized in that: In step 1), the operating conditions of each sensor under each sensing layout scheme are collected, and a total of s fault and normal operating conditions are obtained. The corresponding actual fault and normal operating condition numbers are 1, 2, ... s. There are 200 samples under each operating condition, and a total of 200×s m-dimensional samples are collected, where m is the number of sensors under the current layout. 80% of the data in the 200×s m-dimensional samples are used as the training dataset, and 20% are used as the test dataset. The data collected by each sensor in the training and test datasets are decomposed into three levels using wavelet packet transform, resulting in 8 frequency bands. The energy entropy of the signal in each frequency band after decomposition is calculated as the feature vector of the sensor, forming an 8×m dimensional feature vector.

3. The sensing layout method for fault diagnosis of piston-connecting rod systems according to claim 2, characterized in that: In step 3), the accuracy of fault diagnosis... d represents the number of samples where the predicted operating condition number output during test dataset validation differs from the actual fault and normal operating condition numbers 1, 2, ... s.

4. The sensing layout method for fault diagnosis of piston-connecting rod systems according to claim 1, characterized in that: Step 4) The specific method is: (1) Set the parameters of the binary eagle algorithm, randomly allocate the population size, and generate a set of random n-dimensional binary codes for each individual; (2) Calculate the information entropy, the proportion of sensors deployed in binary codes, and the fitness value of the feature vector corresponding to each randomly generated binary code; (3) Sort all individuals in descending order of fitness value and select the individual with the highest fitness value as the current best position; (4) Update the population position. After updating, normalize the position of each individual in the population. Perform binary encoding mapping on the normalized position and randomly generate a set of n-dimensional [0,1] vectors. In the normalized individual position, if the value of a certain dimension of the normalized individual position is less than the corresponding value of the corresponding dimension of the vector, then the value of that dimension is set and no sensor is installed; otherwise, the sensor is installed. (5) Recalculate the updated fitness value for the individual positions in the mapped population, compare the updated fitness value of the population with the current best fitness value, and if the fitness value in the updated generation is better than the current best fitness value, then take the updated best population individual position and the corresponding fitness value as the current best fitness value and the best population individual position. (6) Determine whether the maximum number of iterations has been reached. After the iteration is completed, the position of the Sky Eagle population is the binary code of the optimal sensor network.

5. A sensing layout method for fault diagnosis of piston-connecting rod systems according to claim 4, characterized in that: In step (2), the information entropy P(z e ) is state z e The probability distribution, z j These are the eigenvalues ​​corresponding to different dimensions of the feature vector, j = 1, 2, ..., 8 × m, where m is the number of sensors in the current layout.

6. The sensing layout method for fault diagnosis of piston and connecting rod systems according to claim 5, characterized in that: The proportion of binary-coded sensors K(2) n -1)=m / n.

7. A sensing layout method for fault diagnosis of piston-connecting rod systems according to claim 6, characterized in that: fitness value fit(2) n -1)=2-F(2 n -1)-G(2 n -1)+K(2 n -1), F(2) n -1) is the fault diagnosis accuracy.

8. A sensing layout method for fault diagnosis of piston-connecting rod systems according to claim 4, characterized in that: In step (4), when updating the population position, the maximum number of iterations is H. If the current iteration number h ≤ 2H / 3, and the random number rand in the range [0,1] ≤ 0.5, the updated individual position is: If rand > 0.5, the updated individual position is: X2(h+1)=X best (h)×Levy(n)+X R (h)+(p-o)×rand, X best (h) represents the optimal solution up to the h-th iteration, X M (h) represents the average position of the population at the h-th iteration, Levy(Dim) is the Levy flight distribution function, and X R (h) is a random solution taken in the range [1, Num] for the h-th iteration, where Num is the population size, and p and o represent the search spiral shape.

9. A sensing layout method for fault diagnosis of piston-connecting rod systems according to claim 4, characterized in that: In step (4), when updating the population position, the maximum number of iterations is H. If the current iteration number h > 2H / 3, and the random number rand in the range [0,1] > 0.5, the updated individual position is: X3(h+1)=(X best (h)-X M (h))×α-rand+((UB-LB)×rand+LB)×α; If rand ≤ 0.5, the updated individual position is: X4(h+1)=QF×X best (h)-(G1×X(h)×rand)-G2×Levy(n)+rand×G1, α is the mining adjustment parameter set to 0.1, QF represents the quality function used to balance the search strategy, G1 represents the various actions a predator takes to track its prey during the hunting process, and G2 represents the flight slope used to follow the prey.

10. A sensing layout method for fault diagnosis of a piston-connecting rod system according to claim 8 or 9, characterized in that: After the update, use the formula The position of each individual in the population is normalized, X b (h+1) represents the individual position updated by the b-th search method in the h+1-th iteration of the Skyhawk algorithm, where b = 1, 2, 3, 4, and the position of each individual is an n-dimensional vector of (0-1).

Images