A method for detecting piston pump soundprint faults

Abstract

The present invention discloses a piston pump voiceprint fault detection method, including: S1, normalizing the voiceprint data sample of the piston pump; S2, performing correlation analysis on the piston pump voiceprint features and fault categories in the voiceprint data sample, and selecting features with strong correlation with the category label; S3, using the above selected features as input and the fault category as the output of the network, constructing the topological structure of the piston pump voiceprint fault detection neural network model, and initializing the network weight and threshold; S4, inputting the training set into the neural network to train the model, and obtaining the training error as the fitness value; S5, initializing the relevant parameters of the whale optimization algorithm HAWOA; S6, using an improved HAWOA based on an adaptive strategy to iteratively search for the best parameters; S7, decoding and assigning it to the neural network framework, and substituting it into the subsequent training to obtain the final piston pump voiceprint fault model. The present invention meets the safety detection needs of chemical production.

Description

Technical Field

[0001] The invention relates to a method for detecting piston pump soundprint faults. Background Art

[0002] A piston pump is a pump device with a fixed seal that sucks in and discharges liquid by moving the piston back and forth. It is widely used in various industrial scenarios for conveying fluids.

[0003] The piston pump relies on the reciprocating motion of the piston to make the working volume of the pump chamber change periodically to achieve the suction and discharge of liquid. During operation, the piston pump will have common fault problems such as wear, corrosion, and blockage.

[0004] Currently, manual inspections or various types of sensor monitoring are usually used to detect faults. However, manual inspections often lead to misjudgments due to lack of personal experience, and sensors can only perform relevant monitoring of monitoring points and cannot fully characterize the operating status of pump equipment, making it difficult to meet the safety testing needs of chemical production. Summary of the invention

[0005] The purpose of the present invention is to provide a piston pump soundprint fault detection method to solve the technical problem of piston pump fault detection in industrial production.

[0006] To this end, the present invention provides a piston pump voiceprint fault detection method, including: S1, normalizing the voiceprint data sample of the piston pump; S2, performing correlation analysis on the piston pump voiceprint features and fault categories in the voiceprint data sample, selecting features with strong correlation with the category label, and dividing them into training set and test set; S3, using the above selected features as input and the fault category as the output of the network, constructing the topological structure of the piston pump voiceprint fault detection neural network model, and initializing the network weight and threshold; S4, inputting the training set into the neural network to train the model, obtaining the training error, and constructing the neural network model error fitness function; S5, setting whale optimization The relevant parameters of the algorithm HAWOA include position vector, population size, maximum number of iterations, and initial convergence factor, wherein the position vector is obtained by encoding the above-mentioned initialized network weights and thresholds; S6, using the error fitness function as the objective function of the HAWOA algorithm, calculating the individual fitness value of the whale according to the objective function, and using an improved HAWOA based on an adaptive strategy to iteratively search for the optimal parameters; S7, decoding the position vector in the optimal parameters to obtain the optimal network weights and thresholds, assigning them to the neural network framework, substituting them into subsequent training to obtain the final piston pump soundprint fault model, and using this model for piston pump soundprint fault detection.

[0007] The beneficial effects of the present invention are as follows:

[0008] The present invention utilizes the voiceprint data of the piston pump operation, and constructs the HAWOA-BP model based on voiceprint recognition and deep learning ideas to improve the efficiency of piston pump fault detection and realize rapid and accurate detection of piston pump faults based on voiceprints. In order to improve the convergence speed and classification accuracy of the BP neural network in piston pump voiceprint fault detection, a BP neural network piston pump voiceprint fault detection model based on an adaptive strategy to improve the whale optimization algorithm is proposed. The weights and thresholds of the whale optimization algorithm are adjusted through an adaptive strategy to enhance the population's global search ability and the ability to jump out of the local optimal solution. The BP neural network uses the optimal weights and thresholds of the improved WOA output as initial parameters to train the model. The present invention is beneficial to promoting piston pump fault detection, realizing rapid and accurate judgment of piston pump operation faults, and further providing a basis for fault diagnosis by monitoring the equipment operation status. It can greatly improve the level of equipment management, maintenance, and economic operation, improve the stability of equipment operation, and reduce the overall operation and maintenance cost. It is convenient and practical.

[0009] In addition to the above-described objects, features and advantages, the present invention has other objects, features and advantages. The present invention will be further described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] The drawings constituting a part of the present application are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0011] Figure 1 To perform data augmentation on the original data;

[0012] Figure 2 is a flowchart of the use of the present invention;

[0013] Figure 3 Execution flowchart for improving the whale optimization algorithm based on adaptive strategy. DETAILED DESCRIPTION

[0014] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0015] Based on the different voiceprint characteristics of the piston pump in normal state and fault state, the present invention uses a neural network model to classify the piston pump state according to the voiceprint data collected during its operation. In order to solve the problem of slow convergence speed and easy to fall into local optimal value during BP neural network training, a new whale optimization algorithm (Whale optimization algorithm based on hybrid adaptive strategy, HAWOA) is proposed to optimize the initial weight and threshold of the BP neural network to achieve a better piston pump voiceprint fault detection effect.

[0016] Please see attached Figure 2 The present invention is a piston pump soundprint fault detection method, which specifically includes the following steps S1-S7.

[0017] S1: Normalize the voiceprint data of the piston pump.

[0018] In order to process the data, the histogram equalization method is used to enhance the data and improve the data quality. The data enhancement process of the original image is shown in the figure. Figure 1 shown.

[0019] S2: Use the Pearson correlation coefficient to perform correlation analysis between the voiceprint features and the piston pump fault categories. Specifically, through correlation analysis, appropriately discard features with weak correlation or irrelevant features, retain features with strong correlation with the category label, and divide the data set into training set and test set in a ratio of 7:3.

[0020] The present invention is mainly aimed at hydraulic piston pumps, and the main fault types are: instantaneous pressure peak fault, blockage or restriction fault, pump housing fatigue fault, etc.; for the convenience of operation, the piston pumps targeted by the present invention belong to the same type, and different types of piston pumps need to retrain the model according to the voiceprint features.

[0021] The Pearson correlation coefficient is used to measure the correlation between two variables X and Y. The calculation formula is as follows:

[0022]

[0023] Among them, ρ X,Y represents the correlation coefficient of X and Y, cov(X,Y) represents the covariance of X and Y, and the calculation formula is E[(X-μ X )(Y-μ Y )],σ X , σ Y Represent the standard deviation of X and Y respectively.

[0024] S3: Initialize the neural network. Specifically, the training set divided in step 2 is used as the input of the neural network, and the fault category is used as the output of the network. The neural network model for piston pump soundprint fault detection is constructed, and the network input nodes and output nodes, maximum number of iterations, learning rate and other model parameters are set according to the number of selected features and the number of fault states, and the connection weights and thresholds are randomly initialized.

[0025] S4: Input the training set into the neural network to train the model, obtain the training error, and use the error as the fitness value. Specifically, use the training set to train the BP neural network, obtain the training error TrainDataErrorRate, use the error as the fitness value, and construct the neural network model error fitness function as follows:

[0026] f=argmin(TrainDataErrorRate)----------------------------------(2)

[0027] S5: HAWOA initialization. Set the relevant parameters of the whale optimization algorithm, encode the neural network weights and thresholds, and obtain the initial population. Specifically, convert the initial weights and thresholds in step 3 into the position vector of HAWOA, set other basic parameters such as population size, maximum number of iterations, initial convergence factor, and randomly initialize the population distribution.

[0028] S6: Use HAWOA to optimize the algorithm. Specifically, the fitness value of step 4 is used as the objective function, and an adaptive strategy is used for optimization until the maximum number of iterations is met or the error accuracy is reached, and the optimal parameters are obtained.

[0029] like Figure 3 As shown, the specific steps of improving the whale optimization algorithm in the present invention are as follows:

[0030] S61: Set formula (2) as the objective function of the whale optimization algorithm, determine the search space dimension according to the number of layers of the neural network and the number of nodes in each layer, set the upper and lower limits of the space lb and ub, the maximum number of iterations T, and initialize the number of iterations t=1.

[0031] S62: Calculate the fitness value of individual whales according to the objective function. In HAWOA, each individual whale in the search space represents a solution, and the optimal individual is That is, the optimal weights and thresholds of the neural network are used as the optimal parameters of the piston pump soundprint fault detection model.

[0032] S63: Simulate the hunting behavior of humpback whale populations, which is mainly divided into two stages: contraction predation and random search. Contraction predation is a local optimal search, encircling prey and spiral update behavior with a probability of 0.5; random search is a global optimization, and the algorithm randomly optimizes in the current feasible solution space.

[0033] Surrounding prey behavior: Assuming the current optimal individual is the target prey, and as the optimal weight and threshold of the neural network, other individuals in the group move to the optimal position. The formula for position update is as follows:

[0034]

[0035]

[0036] in, represents the bracketing step length, is a coefficient vector, defined according to equations (3) and (4). r represents a random number between [0,1]; Represents the linear convergence factor, which decreases between [2,0] as the number of iterations increases.

[0037] Spiral update behavior: Assume that the distance between the i-th whale and the current optimal position is b is a constant coefficient used to limit the logarithmic spiral form; l is a random number between [-1,1], and the position update formula is:

[0038]

[0039] In order to enhance the local search capability, a weight coefficient that can be adaptively changed according to the current population distribution is introduced in the position update behavior in the shrinking predation phase. The formula is as follows:

[0040]

[0041] Among them, I1 and I2 are two constants, set I1 = 10 -4 , I2=10 -4 , P ibest , P iworst Respectively represent the optimal position and the worst position vector in the i-th iteration population; For variable x i The upper and lower bounds of .

[0042] In the early stages of the algorithm iteration, the second part of the adaptive weight The value of is only affected by the number of iterations T i The influence of can maintain a large weight to update the whale position; as the number of iterations increases, the first part of the weight I1·(P ibest -P iworst) can still have a large weight value, causing the whale to jump out of the local optimal solution.

[0043] The probability p is introduced to select the synchronous update of the two behaviors in the shrinking predation stage. The specific formula is as follows:

[0044]

[0045] S64: Random search phase, when |A|≥1, by randomly selecting whales Force it to move away from the reference whale to achieve a global search to obtain the possible location of the prey. The random search formula is as follows:

[0046]

[0047] In order to improve the model's ability to optimize the neural network weights and threshold parameters in the global solution space, an adaptive adjustment threshold is introduced in the random search stage to select the random movement mode of the whale population to enhance the diversity of the population and avoid the entire population moving towards the local optimal individual. The formula is as follows:

[0048]

[0049] in, represents the average fitness value of the population in this iteration, f best 、f worst Respectively represent the best and worst fitness values ​​of the current population.

[0050] For each individual whale, a random value q in [0,1] is compared with the adaptive adjustment threshold H calculated in each round of iteration. If q < H, a whale individual is randomly selected and only the position of the whale is updated according to formula (10), while the positions of other whale individuals remain unchanged. If q > H, the position of the entire whale population is updated according to formula (8).

[0051] X rand =X jlower +r·(X jupper -X jlower )--------------------------(10)

[0052] Among them, r is a random number in [0,1], X jupper , X jlower Represents X rand The upper and lower bounds of the value of .

[0053] S65: Determine whether the termination condition is met. If not, go to S62; otherwise, the algorithm ends and outputs the optimal solution as the optimal weight and threshold of the neural network model.

[0054] S7: Decoding assigns the optimal network weights and thresholds to the BP neural network framework, substitutes them into subsequent training to obtain the HAWOA-BP model, and uses this model for piston pump voiceprint fault detection. Specifically, the optimal weights and thresholds obtained by decoding are assigned to the training model in the BP neural network model to improve the accuracy of the voiceprint detection piston pump fault model.

[0055] The above description is only an embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A piston pump soundprint fault detection method, characterized in that: include: S1. Normalize the voiceprint data sample of the piston pump; S2. Perform correlation analysis on the piston pump voiceprint features and fault categories in the voiceprint data sample, select features with strong correlation with category labels, and divide them into training set and test set; S3. Using the above selected features as input and the fault category as the network output, the topology structure of the piston pump soundprint fault detection neural network model is constructed, and the network weights and thresholds are initialized; S4, input the training set into the neural network to train the model, obtain the training error, and construct the neural network model error fitness function; S5. Setting relevant parameters of the whale optimization algorithm HAWOA, including position vector, population size, maximum number of iterations, and initial convergence factor, wherein the position vector is obtained by encoding the above-mentioned initialization network weight and threshold; S6, using the error fitness function as the objective function of the HAWOA algorithm, calculating the fitness value of the individual whale according to the objective function, and using the HAWOA algorithm to iteratively search for the best parameters; S7. Decode the position vector in the optimal parameters to obtain the optimal network weights and thresholds, assign them to the neural network framework, substitute them into the subsequent training to obtain the final piston pump soundprint fault model, and use this model for piston pump soundprint fault detection. The HAWOA algorithm is an improved whale optimization algorithm based on an adaptive strategy. The iterative optimization using the HAWOA algorithm includes the following steps: S61. Determine the search space dimension according to the number of layers of the neural network and the number of nodes in each layer, and set the upper and lower limits of the space. and , maximum number of iterations , initialize the number of iterations t =1; S62. Each whale individual in the search space represents a solution, the optimal individual are the optimal weights and thresholds of the neural network, which are used as the optimal parameters of the piston pump soundprint fault detection model; S63, simulate the hunting behavior of humpback whale population, which is mainly divided into two stages: contraction predation and random search. Contraction predation is a local optimal search; random search is a global optimization. The algorithm randomly optimizes in the current feasible solution space: Surrounding prey behavior: Assuming the current optimal individual is the target prey, as the optimal weight and threshold of the neural network, Represents the encirclement step length, and all other individuals in the group move to the optimal position. The formula for position update is as follows: -------------------------------------(2) Spiral update behavior: Assume that the distance between the i-th whale and the current optimal position is , b is a constant coefficient used to limit the logarithmic spiral form; is a random number between [-1,1], and the position update formula is: -------------------------------(3) In order to enhance the local search capability, a weight coefficient that can be adaptively changed according to the current population distribution is introduced in the position update behavior of the shrinking predation phase: --------------------------(4) in, , Respectively represent the optimal position and the worst position vector in the i-th iteration population; , For variables The upper and lower bounds of , are two constants, where A and C are coefficient vectors, is the number of iterations, The introduction of probabilistic selection implements the synchronous update of the two behaviors in the shrinking predation stage. The specific formula is as follows: -------------------------(5) S64, random search phase, when By randomly selecting whales , forcing it to move away from the reference whale to achieve global search to obtain the possible location of the prey. The random search formula is as follows: -----------------------------(6) In the random search stage, an adaptive adjustment threshold is introduced to select the random movement mode of the whale population to enhance the diversity of the population. The formula for the adaptive adjustment threshold is as follows: -------------------------------------------------(7) in, Represents the average fitness value of the population in this iteration, , Respectively represent the best and worst fitness values ​​of the current population, For each whale individual, a random value q in [0,1] is used to adjust the adaptive threshold calculated in each iteration. Compare, if , randomly select a whale individual, and according to formula (8), only update the position of the whale, and keep other whale individuals unchanged. , then update the position of the entire whale population according to formula (6): --------------------------------(8) Where r is a random number between 0 and 1. , Respectively represent The upper and lower bounds of the value of S65, determine whether the termination condition is met. If the termination condition is not met, go to S62; otherwise, the algorithm ends and outputs the optimal solution as the optimal weight and threshold of the neural network model.

2. The piston pump soundprint fault detection method according to claim 1, characterized in that: Normalizing the piston pump voiceprint dataset includes data enhancement using the histogram equalization method.

3. The piston pump soundprint fault detection method according to claim 1, characterized in that: The correlation analysis between the piston pump sound print characteristics and the fault category includes: using the Pearson correlation coefficient to perform the correlation analysis between the piston pump sound print characteristics and the piston pump fault category.

4. The piston pump soundprint fault detection method according to claim 1, characterized in that: The decoding in S7 assigns the optimal network weights and thresholds to the BP neural network framework, substitutes them into subsequent training to obtain the HAWOA-BP model, and uses this model for piston pump voiceprint fault detection.

Images