Gear box service life estimation method and device, electronic equipment, storage medium and program product

Abstract

The invention provides a gear box service life estimation method and device, electronic equipment, a storage medium and a program product. The gear box service life estimation method comprises the steps that historical operation signals and working condition information of a target gear box are acquired; performing feature extraction processing on the historical operation signals to obtain operation features; and a trained prediction model is used for processing the operation characteristics and the working condition information, the predicted remaining life is obtained, the prediction model is a machine learning model, and a training sample of the prediction model comprises full-life-cycle operation signals of the gearbox under multiple different working conditions. According to the method, the residual life of the target gearbox can be accurately estimated, so that a scientific basis is provided for optimization of a maintenance strategy of equipment, resource waste is reduced, the maintenance cost is reduced, the monitoring and maintenance efficiency of the gearbox is improved, production stagnation, production accidents and economic losses caused by faults are reduced, and the method is suitable for popularization and application. The method is of great significance in improving the operation efficiency, reliability and safety of mechanical equipment.

Description

Technical Field

[0001] The present disclosure generally relates to the technical field of gearboxes, and more specifically, to a method, device, electronic device, storage medium, and program product for estimating the life of a gearbox. Background Art

[0002] With the development of industrial automation and intelligence, the reliability and safety requirements of mechanical equipment are getting higher and higher. As a key component in mechanical equipment, the health status of the gearbox is directly related to the stability of the entire system.

[0003] Traditional gearbox maintenance strategies are often based on periodic maintenance and lack real-time monitoring and analysis of the actual working status of the gearbox, resulting in inaccurate maintenance decisions, which may cause waste of resources or equipment failure. Summary of the invention

[0004] The present disclosure provides a gearbox life estimation method, device, electronic device, storage medium and program product, which are used to solve at least one of the above problems.

[0005] According to a first aspect of an embodiment of the present disclosure, a gearbox life prediction method is provided, the gearbox life prediction method comprising: obtaining historical operating signals and operating condition information of a target gearbox, wherein the operating signals include vibration signals and / or temperature signals; performing feature extraction processing on the historical operating signals to obtain operating features; using a trained estimation model to process the operating features and the operating condition information to obtain an estimated remaining life, wherein the estimation model is a machine learning model, and the training samples of the estimation model include full life cycle operating signals of the gearbox under multiple different operating conditions.

[0006] Optionally, the estimation model is trained through the following steps: for the estimation model to be trained, randomly determine multiple groups of hyperparameter values ​​to obtain multiple first estimation models, wherein the multiple groups of hyperparameter values ​​correspond one-to-one to the multiple first estimation models; use the training samples to train the multiple first estimation models respectively to obtain multiple second estimation models; use a genetic algorithm to update and train the multiple second estimation models until a preset end condition is met to obtain multiple third estimation models; and determine the trained estimation model from the multiple third estimation models.

[0007] Optionally, the crossover operation in the genetic algorithm includes single-point crossover or double-point crossover; and / or the mutation operation in the genetic algorithm includes Gaussian mutation.

[0008] Optionally, the estimation model is a convolutional neural network, which includes a convolution layer, a pooling layer and a fully connected layer, and the hyperparameters of the estimation model include at least one of the following: the number of convolution kernels, the convolution kernel size, the pooling window size, the number of neurons in the fully connected layer, and the learning rate.

[0009] Optionally, the use of a trained estimation model to process the operating characteristics and the operating condition information to obtain the estimated remaining life includes: inputting the operating characteristics and the operating condition information into the trained estimation model to obtain the estimated remaining life; or inputting the operating characteristics into a sub-model corresponding to the operating condition information in the trained estimation model to obtain the estimated remaining life, wherein the trained estimation model includes multiple sub-models, and the multiple sub-models correspond one-to-one to multiple operating condition information.

[0010] Optionally, the operating characteristics include at least one or more combinations of the following: time domain characteristics, frequency domain characteristics, and time-frequency domain characteristics, wherein the time domain characteristics include at least one or more combinations of the following: peak value, mean value, variance, skewness, maximum value, minimum value, root mean square value, and waveform coefficient, the frequency domain characteristics include power spectral density, and the time-frequency domain characteristics include wavelet transform coefficients.

[0011] According to a second aspect of an embodiment of the present disclosure, a gearbox life prediction device is provided, comprising: an acquisition unit, configured to acquire historical operating signals and operating condition information of a target gearbox, wherein the operating signals include vibration signals and / or temperature signals; an extraction unit, configured to perform feature extraction processing on the historical operating signals to obtain operating features; and an estimation unit, configured to use a trained estimation model to process the operating features and the operating condition information to obtain an estimated remaining life, wherein the estimation model is a machine learning model, and the training samples of the estimation model include full life cycle operating signals of the gearbox under multiple different operating conditions.

[0012] Optionally, the estimation model is trained through the following steps: for the estimation model to be trained, randomly determine multiple groups of hyperparameter values ​​to obtain multiple first estimation models, wherein the multiple groups of hyperparameter values ​​correspond one-to-one to the multiple first estimation models; use the training samples to train the multiple first estimation models respectively to obtain multiple second estimation models; use a genetic algorithm to update and train the multiple second estimation models until a preset end condition is met to obtain multiple third estimation models; and determine the trained estimation model from the multiple third estimation models.

[0013] Optionally, the crossover operation in the genetic algorithm includes single-point crossover or double-point crossover; and / or the mutation operation in the genetic algorithm includes Gaussian mutation.

[0014] Optionally, the estimation model is a convolutional neural network, which includes a convolution layer, a pooling layer and a fully connected layer, and the hyperparameters of the estimation model include at least one of the following: the number of convolution kernels, the convolution kernel size, the pooling window size, the number of neurons in the fully connected layer, and the learning rate.

[0015] Optionally, the estimation unit is further configured to: input the operating characteristics and the operating condition information into the trained estimation model to obtain the estimated remaining life; or input the operating characteristics into a sub-model corresponding to the operating condition information in the trained estimation model to obtain the estimated remaining life, wherein the trained estimation model includes multiple sub-models, and the multiple sub-models correspond one-to-one to multiple operating condition information.

[0016] Optionally, the operating characteristics include at least one or more combinations of the following: time domain characteristics, frequency domain characteristics, and time-frequency domain characteristics, wherein the time domain characteristics include at least one or more combinations of the following: peak value, mean value, variance, skewness, maximum value, minimum value, root mean square value, and waveform coefficient, the frequency domain characteristics include power spectral density, and the time-frequency domain characteristics include wavelet transform coefficients.

[0017] According to a third aspect of an embodiment of the present disclosure, an electronic device is provided, comprising: at least one processor; and at least one memory storing computer executable instructions, wherein the computer executable instructions, when executed by the at least one processor, prompt the at least one processor to execute a gearbox life prediction method according to an exemplary embodiment of the present disclosure.

[0018] According to a fourth aspect of an embodiment of the present disclosure, a computer-readable storage medium is provided. When instructions in the computer-readable storage medium are executed by at least one processor, the at least one processor is prompted to execute a gearbox life prediction method according to an exemplary embodiment of the present disclosure.

[0019] According to a fifth aspect of an embodiment of the present disclosure, a computer program product is provided, comprising computer instructions, which, when executed by at least one processor, prompt the at least one processor to execute a gearbox life prediction method according to an exemplary embodiment of the present disclosure.

[0020] The technical solution provided by the embodiments of the present disclosure brings at least the following beneficial effects: according to the gearbox life prediction method and device, electronic device, and storage medium disclosed in the present disclosure, by analyzing data such as historical operating signals and operating condition information of the target gearbox, the remaining life of the target gearbox can be accurately estimated, and its working status can be monitored in real time based on the working process data of the target gearbox, and potential failure risks can be discovered in time, thereby providing a scientific basis for optimizing the maintenance strategy of the equipment, reducing resource waste, lowering maintenance costs, and improving the monitoring and maintenance efficiency of the gearbox, and helping to reduce production stagnation, production accidents, and economic losses caused by failures, which is of great significance to improving the operating efficiency, reliability, and safety of mechanical equipment.

[0021] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The drawings herein are incorporated into and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the description are used to explain the principles of the present disclosure, and do not constitute improper limitations on the present disclosure.

[0023] Figure 1 is a flow chart of a gearbox life prediction method according to an exemplary embodiment of the present disclosure.

[0024] Figure 2 It is a technical roadmap of a gearbox life prediction method according to a specific embodiment of the present disclosure.

[0025] Figure 3 It is a structural diagram of an estimation model according to a specific embodiment of the present disclosure.

[0026] Figure 4 is a block diagram of a gearbox life prediction apparatus according to an exemplary embodiment of the present disclosure.

[0027] Figure 5 is a block diagram of an electronic device according to an exemplary embodiment of the present disclosure. DETAILED DESCRIPTION

[0028] In order to enable ordinary persons in the art to better understand the technical solutions of the present disclosure, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the accompanying drawings.

[0029] It should be noted that the terms "first", "second", etc. in the specification and claims of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. The implementation methods described in the following examples do not represent all implementation methods consistent with the present disclosure. Instead, they are merely examples of devices and methods consistent with some aspects of the present disclosure as detailed in the attached claims.

[0030] It should be noted that the phrase "at least one of the items" in the present disclosure includes three types of parallel situations: "any one of the items", "a combination of any number of the items", and "all of the items". For example, "including at least one of A and B" includes the following three parallel situations: (1) including A; (2) including B; (3) including A and B. Another example is "executing at least one of step 1 and step 2" which means the following three parallel situations: (1) executing step 1; (2) executing step 2; (3) executing step 1 and step 2.

[0031] Hereinafter, a gearbox life prediction method and apparatus, an electronic device, and a storage medium according to exemplary embodiments of the present disclosure will be described in detail with reference to the accompanying drawings.

[0032] Figure 1 1 is a flow chart of a method for predicting the life of a gearbox according to an exemplary embodiment of the present disclosure. The method can be executed on an electronic device with sufficient computing power.

[0033] Reference Figure 1 In step S101, historical operation signals and operating condition information of the target gearbox are obtained.

[0034] The operating signal includes a vibration signal and / or a temperature signal, and may also include other signals for reflecting the operating conditions of the gearbox.

[0035] In terms of time, the historical operation signal may specifically include all the operation signals since the target gearbox was put into use, and may also include the operation signal of the target gearbox within a specified period, and the life estimation is realized in a corresponding manner. For example, for the former, the life consumed by the target gearbox since it was put into use can be estimated using the estimation model introduced later, so as to subtract the estimated consumed life from the design life to obtain the estimated remaining life, or the estimation model introduced later can be directly used to obtain the estimated remaining life. For the latter, the remaining life of the target gearbox at the starting point of the specified period can also be obtained (for example, including but not limited to the design life minus the actual running time), and the life consumed in the specified period can be estimated using the estimation model introduced later, so as to subtract the estimated consumed life in the specified period from the remaining life at the starting point of the specified period to obtain the estimated remaining life, or the remaining life of the target gearbox at the starting point of the specified period or the actual running time can be used as the input data of the estimation model introduced later, and the estimated remaining life can be directly obtained using the estimation model. The present disclosure is not limited to this.

[0036] The operating condition information is used to describe the operating condition of the target gearbox, including but not limited to load, speed, temperature, lubrication conditions, etc. Different operating conditions have different consumption on the equipment. By obtaining the operating condition information, a more comprehensive information reference can be provided for the estimation of the remaining life.

[0037] In step S102, feature extraction processing is performed on the historical operation signal to obtain operation features.

[0038] As an example, the feature extraction process may specifically include preprocessing and feature extraction, and the preprocessing may include but is not limited to filtering, denoising, etc. to improve signal quality.

[0039] Optionally, the operating characteristics include at least one or more of the following combinations: time domain characteristics, frequency domain characteristics, time-frequency domain characteristics, wherein the time domain characteristics include at least one or more of the following combinations: peak value, mean value, variance, skewness, maximum value, minimum value, root mean square value and waveform coefficient, the frequency domain characteristics include power spectrum density, and the time-frequency domain characteristics include wavelet transform coefficients. By extracting rich operating characteristics of different domains from historical operating signals, the operating conditions of the target gearbox can be fully described from different angles, thereby reflecting the characteristic differences of different remaining lifespans and providing a reliable basis for the estimation of remaining lifespans. It should be understood that for embodiments in which the operating signal includes more than two signals, features need to be extracted for each signal separately. For example, for embodiments that include both vibration signals and temperature signals, vibration features and temperature features need to be extracted separately. Specifically, the peak value represents the local maximum value. All operating signals can be divided into several segments, and the maximum value of each segment or specified segment is obtained as the peak value, and the maximum value is the maximum value in the global range (i.e., all historical operating signals). The waveform coefficient is a dimensionless quantity in an AC signal, which is used to represent the ratio of the root mean square value of the signal to the rectified average.

[0040] In step S103, the trained estimation model is used to process the operation characteristics and operating condition information to obtain the estimated remaining life.

[0041] The estimation model is a machine learning model, such as but not limited to support vector machines, convolutional neural networks, etc. The training samples of the estimation model include the full life cycle operation signals of the gearbox under multiple different working conditions. It should be understood that the full life cycle operation signal refers to the operation signal of the entire cycle from the start of the use of the gearbox to the scrapping of the gearbox, which can determine the accurate remaining life of the gearbox as a sample at any time, thereby obtaining comprehensive and reliable training sample data. It should also be understood that the gearbox itself will often experience different working conditions during its full life cycle, thereby obtaining operating signals under different working conditions. At the same time, the difference in working conditions of the same gearbox is often not too large. Therefore, in order to ensure the richness of the samples, different gearboxes with large overall differences in working conditions can also be actively selected to obtain training samples.

[0042] Regarding the use of the operating condition information in step S103, in some embodiments, optionally, step S103 includes: inputting the operating characteristics and operating condition information into a trained estimation model to obtain an estimated remaining life. By directly using the operating condition information as input data for the estimation model, that is, using the same estimation model for different operating conditions, the generalization ability of the model can be improved and the life estimation operation can be simplified. It should be understood that the description here as "inputting" the operating characteristics and operating condition information into the trained estimation model is to distinguish it from the original description of step S103 above, and does not limit the estimation model to only directly process the operating characteristics and operating condition information. As an example, the operating condition characteristics can also be extracted from the operating condition information first, and the estimation model processes the operating characteristics and operating condition characteristics. The present disclosure does not limit this.

[0043] In other embodiments, optionally, step S103 includes: inputting the operating characteristics into the sub-model corresponding to the operating condition information in the trained estimation model to obtain the estimated remaining life, wherein the trained estimation model includes multiple sub-models, and the multiple sub-models correspond one-to-one to the multiple operating condition information. By training the corresponding sub-models for different operating conditions, more detailed feature differences can be mined from a single operating condition, and a more refined life estimation can be achieved, which helps to improve the accuracy of life estimation. As an example, since the operating condition information can be represented by the values ​​of specific variables (such as but not limited to the load, speed, temperature, lubrication conditions, etc. introduced above), different operating conditions can be divided according to different value ranges of specific variables. Further, for the case where the operating condition information is represented by the values ​​of multiple specific variables, different operating conditions can be divided according to the combination of the value ranges of each variable. The present disclosure does not limit the specific value range division and combination method.

[0044] According to the gearbox life prediction method disclosed in the present invention, by analyzing data such as historical operating signals and operating condition information of the target gearbox, the remaining life of the target gearbox can be accurately estimated, and its working status can be monitored in real time based on the working process data of the target gearbox, so as to promptly discover potential failure risks (for example, including but not limited to taking the difference between the design life of the target gearbox and the operating time as the theoretical remaining life, and considering that there is a failure risk when the estimated remaining life is significantly less than the theoretical remaining life), thereby providing a scientific basis for optimizing the maintenance strategy of the equipment, reducing resource waste, reducing maintenance costs, and improving the monitoring and maintenance efficiency of the gearbox, and helping to reduce production stagnation, production accidents and economic losses caused by failures, which is of great significance for improving the operating efficiency, reliability and safety of mechanical equipment.

[0045] As an example, the estimated remaining life obtained can be used as a basis for formulating a maintenance strategy, which includes, but is not limited to, a replacement time of the gearbox, a maintenance cycle, and a spare parts ordering plan.

[0046] Next, the training of the estimation model is introduced.

[0047] Optionally, the estimation model is trained by the following steps: for the estimation model to be trained, a plurality of sets of hyperparameter values ​​are randomly determined to obtain a plurality of first estimation models, wherein the plurality of sets of hyperparameter values ​​correspond one-to-one to the plurality of first estimation models; using training samples, the plurality of first estimation models are respectively trained to obtain a plurality of second estimation models; using a genetic algorithm, the plurality of second estimation models are subjected to hyperparameter update and training until a preset termination condition is met to obtain a plurality of third estimation models; and a trained estimation model is determined from the plurality of third estimation models. By using a genetic algorithm to update the hyperparameters of the estimation model and training the obtained estimation model during the updating process, it is possible to realize automatic optimization search of the hyperparameters of the estimation model, which helps to improve the performance of the estimation model.

[0048] Specifically, we can first determine the multiple hyperparameters to be updated to form an individual vector as the search target of the genetic algorithm. Then, we randomly generate a group of initial individuals as the first generation of individuals in the population. Each initial individual corresponds to a group of hyperparameter values. Each hyperparameter value is randomly determined within a predetermined range of values. Each initial individual also corresponds to a first estimation model. Next, we use these initial individuals as the starting point and repeatedly perform cyclic calculations to simulate the inheritance of genes from generation to generation.

[0049] Each cycle calculation includes selection operation, crossover operation and mutation operation. The selection operation is used to select excellent individuals from a current group of individuals as the parents of the next generation of individuals. Taking the first cycle calculation as an example, when the selection operation is performed, the first estimation model corresponding to these individuals is first trained to adjust the model parameters to obtain multiple second estimation models. The fitness values ​​of these second estimation models are then calculated (the fitness calculation method is determined in advance, for example, it can be consistent with the loss function used in model training, and it can be consistent with the evaluation function used in model testing. Of course, other reasonable calculation methods can also be used, and the present disclosure is not limited to this), as the fitness values ​​of the corresponding initial individuals. Then, several individuals with relatively good fitness values ​​are selected from each initial individual (for example, when the fitness is consistent with the loss function, several individuals with the smallest fitness value can be selected, that is, each initial individual is arranged in descending / ascending order according to the fitness value, and the last / first individuals are selected) as the parents of the next generation of individuals. The crossover operation is used to exchange chromosomes of the selected parent individuals, that is, to exchange hyperparameter values, to imitate the gene crossover of two individuals in heredity. The mutation operation is used to perform genetic mutations on the individuals obtained by crossover, that is, to make small changes to some chromosomes of some crossover individuals, imitating the genetic mutations of individuals in genetics to increase the diversity of the population.

[0050] After one cycle of calculation, the next generation of individuals can be obtained. Then continue to perform a new round of cycle calculation on the next generation of individuals, and inherit from generation to generation until the preset end conditions are met, such as but not limited to the number of cycles reaching a predetermined value and the fitness value convergence. The model corresponding to the generation of individuals finally obtained is a plurality of third estimation models, and one with the best fitness value is selected from these third estimation models as the trained estimation model. As an example, if the fitness function contains at least two sub-items with opposite correlations, for example, the smaller the value of one of the sub-items, the higher the fitness, and the larger the value of the other sub-item, the higher the fitness, then these items with opposite correlations can be reasonably numerically processed, such as but not limited to adding a negative coefficient to one of the sub-items, such as -1, so that the correlations of all sub-items of the fitness function are the same, and the present disclosure does not limit this.

[0051] Optionally, the crossover operation in the genetic algorithm includes a single-point crossover or a double-point crossover. Single-point crossover refers to determining a fixed crossover point in the individual vector, such as but not limited to the midpoint of the individual vector, so that the individual vector is divided into two parts, front and back, with the crossover point as the boundary, and the individual vectors of the two parent generations are partially exchanged, that is, the first half vector of the first parent generation and the second half vector of the second parent generation form an individual vector of a child generation, and the first half vector of the second parent generation and the second half vector of the first parent generation form an individual vector of another child generation. Double-point crossover refers to determining two fixed crossover points in the individual vector, so that the individual vector is divided into three parts with two crossover points as the boundary, and the middle parts of the individual vectors of the two parent generations are exchanged with each other to obtain the individual vectors of the two child generations. By implementing the crossover operation by adopting a single-point crossover or a double-point crossover method, the efficiency of the crossover operation can be improved, and it is also helpful to maintain certain continuous parts of the parent chromosome.

[0052] Optionally, the mutation operation in the genetic algorithm includes Gaussian mutation. Gaussian mutation refers to adding a random number that conforms to the Gaussian distribution (called Gaussian noise) to the individual vector. By adopting Gaussian mutation, a more detailed search can be performed in the local area. As an example, the mutation operation can be performed on a specified part of the hyperparameters, or one or more hyperparameters can be randomly selected for mutation operation, and Gaussian operation can also be performed on all hyperparameters, and the present disclosure does not limit this.

[0053] Optionally, the estimation model is a convolutional neural network, and the estimation model includes a convolutional layer, a pooling layer and a fully connected layer, and the hyperparameters of the estimation model include at least one of the following: the number of convolution kernels, the size of the convolution kernel, the size of the pooling window, the number of neurons in the fully connected layer, and the learning rate. By using a convolutional neural network as an estimation model, data such as frequency, interval, and duration can be efficiently processed. Through the convolution operation, local features of frequency, interval, and duration can be extracted, and parameter sharing can be used to significantly reduce the number of model parameters, thereby reducing computational complexity. The introduction of the pooling layer not only reduces the data dimension, but also enhances the robustness of the model to changes such as translation and rotation. In addition, the hierarchical structure of the convolutional neural network enables it to learn low-level features from the shallow layer to high-level features from the deep layer, thereby improving the model's ability to learn features. By genetically updating at least one of the above hyperparameters, it is helpful to search for the optimal convolutional neural network structure, further improving the model performance.

[0054] Next, combine Figure 2 A method for estimating the life of a gearbox according to a specific embodiment of the present disclosure is introduced.

[0055] The operation signal used in this specific embodiment is specifically a vibration signal. Of course, as mentioned above, in other embodiments, other signals can also be reasonably selected as the operation signal. Figure 2 ,This specific embodiment includes two parts of model training and reasoning, which mainly includes the following five steps, of which the first four steps are the training part, and the last step is the reasoning part.

[0056] Step 1: Collect the full life cycle vibration signal of the high-speed gearbox under different working conditions, and set the sampling frequency to f s =25kHz, sampling interval T s =0.5min, each sampling time is T d =1.5s.

[0057] Step 2: preprocess the collected vibration signal x(t), including filtering, denoising, etc., to improve signal quality.

[0058] As an example, a low-pass filter is used to remove high-frequency noise. Its transfer function can be expressed as:

[0059]

[0060] Where f is the signal frequency, f c is the cutoff frequency. The filtered vibration signal x f (t) can be obtained by inverse Fourier transform:

[0061]

[0062] Where X(f) is the Fourier transform of the original vibration signal x(t), Represents the inverse Fourier transform.

[0063] Step 3: From the filtered and denoised vibration signal x f (t) Extract vibration features, including the time domain features A of the vibration signal time , frequency domain features A freq and time-frequency domain features A time-freq .

[0064] Time domain characteristics of vibration signals A time Including Peak A peak , mean A mean , Variance A var , skewness A skew , defined as follows:

[0065] A peak =max(x f (t));

[0066]

[0067] Where T represents the duration of the full life cycle, and the vibration signal x during the sampling interval is f (t) takes 0.

[0068] Frequency domain characteristics of vibration signal A freq It can be obtained by Fast Fourier Transform (FFT), including the power spectrum density P(f), which is defined as follows:

[0069]

[0070] Time-frequency domain characteristics of vibration signals time-freq Including the wavelet transform coefficients W(a,b), defined as follows:

[0071]

[0072] Among them, ψ(t) is the wavelet function, * indicates the complex conjugate operation on the parameter, a is the translation parameter, and b is the scale parameter. Regarding the values ​​of the translation parameter a and the scale parameter b, if a specific time period of the signal needs to be analyzed, a can be set to the center point of the time period; for the analysis of low-frequency components, a larger b value can be selected, while for the analysis of high-frequency components, a smaller b value can be selected.

[0073] Step 4: Use convolutional neural network based on the extracted time domain feature A time , frequency domain features A freq and time-frequency domain features A time-freq Train the estimation model. Figure 3 As shown, firstly, three 128×128 pixel features (corresponding to the time domain features A time , frequency domain features A freq and time-frequency domain features A time-freq ) is input into the input layer of this convolutional neural network, which is equipped with 3 channels, carrying the above three features respectively. After being processed by the input layer, the features of each channel can be extracted to obtain 8 feature maps of 64×64 pixels. The next layer of the network is the first convolution layer. After the convolution operation, 24 feature maps of 48×48 pixels are obtained. Each feature map is the result of the convolution operation between the input feature map and a convolution kernel. It is followed by the pooling layer, whose main function is to reduce the spatial size of the feature map. This helps to reduce the number of parameters and computational complexity of subsequent layers, and at the same time improves the translation invariance of the features. After the pooling operation, the size of each feature map will be reduced to 16×16 pixels. Finally, the network integrates the features extracted by the previous convolution layer and pooling layer through a fully connected layer to obtain a 1×256 pixel feature map for the final classification or regression task. In this specific embodiment, it is specifically used to estimate the remaining life of the high-speed gearbox.

[0074] The loss function L used in the training of the estimation model is defined as:

[0075]

[0076] Among them, y i is the actual remaining life, is the estimated remaining life output by the model, and N is the number of samples.

[0077] At the output layer, the gradient of the loss function with respect to the network output is calculated, i.e.

[0078] The chain rule allows us to calculate the gradient of the loss function with respect to the network parameters. For the activation function σ, we first calculate the gradient of the loss function with respect to the output of the activation function, and then multiply it by the derivative of the activation function (i.e. σ′) to get the gradient of the loss function with respect to the weights and biases:

[0079]

[0080] Among them, W is the weight matrix, b is the bias term, and x is the input feature, including the time domain feature A time , frequency domain features A freq and time-frequency domain features A time-freq .

[0081] For the hidden layers, we need to recursively calculate the gradient of the loss function with respect to the weights and biases of each layer using the chain rule. For the lth layer, the gradient is calculated as follows:

[0082]

[0083] Among them, a (l) is the activation output of layer l, W (l) and b (l) are the weight and bias of the lth layer respectively.

[0084] Once the gradient is calculated, you can use gradient descent or its variants, such as but not limited to Adam (Adaptive Moment Estimation), RMSprop (Root Mean Square Propagation), etc., to update the weights and biases of the network:

[0085]

[0086] Where η is the learning rate. This process is repeated on the entire training set until the network weights and biases converge, or the preset number of iterations is reached.

[0087] In convolutional neural networks, the backpropagation process is similar to that of fully connected networks, but with additional considerations for weight sharing in convolutional layers and the characteristics of pooling layers. In convolutional layers, the calculation of gradients needs to take into account the convolution operation, while in pooling layers, the propagation of gradients needs to take into account the characteristics of the pooling operation. Backpropagation for these layers involves convolution and upsampling operations on the input feature maps to calculate the gradient of the loss function with respect to the convolution kernel and pooling parameters.

[0088] After training, the mean absolute error (MAE) and the coefficient of determination (R 2 ) is used as the evaluation function to evaluate the performance of the trained model.

[0089] The mean absolute error (MAE) is calculated as follows:

[0090]

[0091] Coefficient of determination (R 2 ) is calculated as follows:

[0092]

[0093] in, is the average of the actual remaining life of each sample in the test set.

[0094] In the life prediction task, the optimization objectives are to minimize the mean absolute error and maximize the coefficient of determination.

[0095] The present disclosure also uses a genetic algorithm to search for the optimal hyperparameters or structure of a convolutional neural network to improve model performance.

[0096] Specifically, the hyperparameters or structures of the convolutional neural network are encoded as chromosomes (individuals). The hyperparameters that need to be optimized include: the number of convolution kernels C, the pooling window size P×P, the convolution kernel size K×K, the number of neurons in the fully connected layer F, and the learning rate η.

[0097] Each individual is represented by a vector: individual = [C, P, K, F, η].

[0098] Then, a group of initial individuals (population) is randomly generated, and the population size is N pop The hyperparameters in each individual are randomly initialized within the preset value range.

[0099] Each individual (hyperparameter value combination) is trained and evaluated, and its fitness value is calculated. The fitness function is the mean absolute error and the coefficient of determination used on the test set.

[0100] Select excellent individuals to enter the next generation based on their fitness values.

[0101] Perform a crossover operation on the selected individuals to generate new individuals. The crossover method is a two-point crossover, specifically, a crossover point is determined after the second hyperparameter and the fourth hyperparameter, which can be expressed as:

[0102] Offspring 1 = (C and P of parent 1) + (K and F of parent 2) + (η of parent 1);

[0103] Offspring 2 = (C and P of parent 2) + (K and F of parent 1) + (η of parent 2).

[0104] Mutation operations are performed on some individuals obtained by crossover to increase the diversity of the population. The mutation method is Gaussian mutation, that is, adding Gaussian noise to the gene value, which can be expressed as:

[0105] C′=C+N(0,σ 2 ).

[0106] Among them, N(0,σ 2 ) means the mean is 0 and the variance is σ 2 It should be understood that σ here represents the variance of the Gaussian distribution, not the activation function introduced above.

[0107] The above selection, crossover and mutation operations are repeated until the predetermined number of iterations T is reached and the fitness value converges. Finally, the individual with the best fitness is selected as the optimal estimation model configuration.

[0108] By optimizing the estimation model through genetic algorithms, the optimal model structure and hyperparameters can be automatically searched to improve the performance of the estimation model.

[0109] Step 5: Apply the trained model to new vibration signal data to estimate the remaining life of the gearbox.

[0110] Figure 4 is a block diagram of a gearbox life prediction device according to an exemplary embodiment of the present disclosure. Figure 4 The gearbox life estimation device 400 includes an acquisition unit 401 , an extraction unit 402 , and an estimation unit 403 .

[0111] The acquisition unit 401 may acquire historical operating signals and operating condition information of the target gearbox, wherein the operating signals include vibration signals and / or temperature signals.

[0112] The extraction unit 402 may perform feature extraction processing on the historical operation signal to obtain the operation feature.

[0113] The estimation unit 403 can use a trained estimation model to process operating characteristics and operating condition information to obtain an estimated remaining life, wherein the estimation model is a machine learning model, and the training samples of the estimation model include full life cycle operating signals of the gearbox under multiple different operating conditions.

[0114] Optionally, the estimation model is trained through the following steps: for the estimation model to be trained, randomly determine multiple groups of hyperparameter values ​​to obtain multiple first estimation models, wherein the multiple groups of hyperparameter values ​​correspond one-to-one to the multiple first estimation models; use training samples to train the multiple first estimation models separately to obtain multiple second estimation models; use a genetic algorithm to update and train the hyperparameters of the multiple second estimation models until preset end conditions are met to obtain multiple third estimation models; and determine a trained estimation model from the multiple third estimation models.

[0115] Optionally, the crossover operation in the genetic algorithm includes one-point crossover or two-point crossover; and / or the mutation operation in the genetic algorithm includes Gaussian mutation.

[0116] Optionally, the estimation model is a convolutional neural network, which includes a convolution layer, a pooling layer and a fully connected layer. The hyperparameters of the estimation model include at least one of the following: the number of convolution kernels, the size of the convolution kernel, the size of the pooling window, the number of neurons in the fully connected layer, and the learning rate.

[0117] Optionally, the estimation unit 403 may also: input the operating characteristics and operating condition information into a trained estimation model to obtain an estimated remaining life; or input the operating characteristics into a sub-model corresponding to the operating condition information in the trained estimation model to obtain an estimated remaining life, wherein the trained estimation model includes multiple sub-models, and the multiple sub-models correspond one-to-one to the multiple operating condition information.

[0118] Optionally, the operating characteristics include at least one or more combinations of: time domain characteristics, frequency domain characteristics, and time-frequency domain characteristics, wherein the time domain characteristics include at least one or more combinations of: peak, mean, variance, skewness, maximum value, minimum value, root mean square value, and waveform coefficient, the frequency domain characteristics include power spectral density, and the time-frequency domain characteristics include wavelet transform coefficients.

[0119] Regarding the device in the above embodiment, the specific manner in which each unit performs the operation has been described in detail in the embodiment of the method, and will not be elaborated here.

[0120] Figure 5 A structural block diagram of an electronic device 500 according to an exemplary embodiment of the present disclosure is shown.

[0121] Reference Figure 5 The electronic device 500 includes: at least one memory 501 and at least one processor 502, wherein the at least one memory 501 stores computer executable instructions, and when the computer executable instructions are executed by the at least one processor 502, the at least one processor is prompted to execute the target corresponding method as described in the above exemplary embodiment.

[0122] As an example, the electronic device 500 may be a PC, a tablet device, a personal digital assistant, a smart phone, or other device capable of executing the above instruction set. Here, the electronic device 500 is not necessarily a single electronic device 500, but may also be any device or circuit capable of executing the above instruction (or instruction set) individually or in combination. The electronic device 500 may also be part of an integrated control system or system manager, or may be configured as a portable electronic device 500 that is interconnected with a local or remote (e.g., via wireless transmission) interface.

[0123] In the electronic device 500, the processor 502 may include a central processing unit (CPU), a graphics processing unit (GPU), a programmable logic device, a dedicated processor system, a microcontroller or a microprocessor. As an example and not limitation, the processor 502 may also include an analog processor, a digital processor, a microprocessor, a multi-core processor, a processor array, a network processor, etc.

[0124] The processor 502 may execute instructions or codes stored in the memory 501, wherein the memory 501 may also store data. Instructions and data may also be sent and received over a network via a network interface device, wherein the network interface device may employ any known transmission protocol.

[0125] The memory 501 may be integrated with the processor 502, for example, by placing RAM or flash memory within an integrated circuit microprocessor or the like. In addition, the memory 501 may include a separate device, such as an external disk drive, a storage array, or any other storage device that can be used by a database system. The memory 501 and the processor 502 may be operatively coupled, or may communicate with each other, such as through an I / O port, a network connection, etc., so that the processor 502 can read files stored in the memory.

[0126] In addition, the electronic device 500 may also include a video display (such as a liquid crystal display) and a user interaction interface (such as a keyboard, a mouse, a touch input device, etc.) All components of the electronic device 500 may be connected to each other via a bus and / or a network.

[0127] According to an exemplary embodiment of the present disclosure, a computer-readable storage medium storing instructions may also be provided, wherein the instructions, when executed by at least one processor, cause the at least one processor to execute the target corresponding method as described in the above exemplary embodiment. Examples of computer-readable storage media here include: read-only memory (ROM), random access programmable read-only memory (PROM), electrically erasable programmable read-only memory (EEPROM), random access memory (RAM), dynamic random access memory (DRAM), static random access memory (SRAM), flash memory, non-volatile memory, CD-ROM, CD-R, CD+R, CD-RW, CD+RW, DVD-ROM, DVD-R, DVD+R, DVD-RW, DVD+RW, DVD-RAM, BD-ROM, BD-R, BD-R LTH, BD-RE, Blu-ray or optical disk storage, hard disk drive (HDD), solid state drive (SSD), card storage (such as, multimedia card, secure digital (SD) card or extreme digital (XD) card), magnetic tape, floppy disk, magneto-optical data storage device, optical data storage device, hard disk, solid state disk and any other device, any other device is configured to store computer programs and any associated data, data files and data structures in a non-transitory manner and provide the computer programs and any associated data, data files and data structures to a processor or computer so that the processor or computer can execute the computer program. The computer program in the above-mentioned computer-readable storage medium can be run in an environment deployed in a computer device such as a client, a host, an agent device, a server, etc. In addition, in one example, the computer program and any associated data, data files and data structures are distributed on a networked computer system, so that the computer program and any associated data, data files and data structures are stored, accessed and executed in a distributed manner by one or more processors or computers.

[0128] According to an exemplary embodiment of the present disclosure, a computer program product may also be provided, including computer instructions, which, when executed by at least one processor, execute the target corresponding method as described in the above exemplary embodiment.

[0129] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. The present disclosure is intended to cover any variations, uses or adaptations of the present disclosure, which follow the general principles of the present disclosure and include common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The description and examples are intended to be exemplary only, and the true scope and spirit of the present disclosure are indicated by the appended claims.

[0130] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A method for estimating the life of a gearbox, characterized in that: The gearbox life estimation method comprises: Acquire historical operating signals and operating condition information of a target gearbox, wherein the operating signals include vibration signals and / or temperature signals; Performing feature extraction processing on the historical operation signal to obtain operation features; Use the trained estimation model to process the operating characteristics and the operating condition information to obtain an estimated remaining life. Wherein, the estimation model is a machine learning model, and the training samples of the estimation model include full life cycle operation signals of the gearbox under multiple different working conditions.

2. The gearbox life prediction method according to claim 1, characterized in that: The estimation model is trained by the following steps: For the estimation model to be trained, randomly determine multiple sets of hyperparameter values ​​to obtain multiple first estimation models, wherein the multiple sets of hyperparameter values ​​correspond one-to-one to the multiple first estimation models; Using the training samples, respectively training the multiple first estimation models to obtain multiple second estimation models; Using a genetic algorithm, performing hyperparameter update and training on the plurality of second estimation models until a preset end condition is met, thereby obtaining a plurality of third estimation models; The trained estimation model is determined from the multiple third estimation models.

3. The gearbox life prediction method according to claim 2, characterized in that: The crossover operation in the genetic algorithm includes a single-point crossover or a double-point crossover; and / or The mutation operation in the genetic algorithm includes Gaussian mutation.

4. The gearbox life prediction method according to claim 2, characterized in that: The estimation model is a convolutional neural network, which includes a convolution layer, a pooling layer and a fully connected layer. The hyperparameters of the estimation model include at least one of the following: the number of convolution kernels, the size of the convolution kernel, the size of the pooling window, the number of neurons in the fully connected layer, and the learning rate.

5. The gearbox life prediction method according to any one of claims 1 to 4, characterized in that: The using the trained estimation model to process the operating characteristics and the operating condition information to obtain the estimated remaining life includes: Inputting the operating characteristics and the operating condition information into the trained estimation model to obtain the estimated remaining life; or The operating characteristics are input into the sub-model corresponding to the operating condition information in the trained estimation model to obtain the estimated remaining life, wherein the trained estimation model includes multiple sub-models, and the multiple sub-models correspond one-to-one to multiple operating condition information.

6. The method for estimating the life of a gearbox according to any one of claims 1 to 4, characterized in that: The operating characteristics include at least one or more combinations of the following: time domain characteristics, frequency domain characteristics, and time-frequency domain characteristics, wherein the time domain characteristics include at least one or more combinations of the following: peak value, mean value, variance, skewness, maximum value, minimum value, root mean square value, and waveform coefficient; the frequency domain characteristics include power spectral density; and the time-frequency domain characteristics include wavelet transform coefficients.

7. A gearbox life prediction device, characterized in that: The gearbox life prediction device comprises: an acquisition unit, configured to acquire historical operation signals and operating condition information of a target gearbox, wherein the operation signal includes a vibration signal and / or a temperature signal; an extraction unit, configured to perform feature extraction processing on the historical operation signal to obtain an operation feature; an estimation unit, configured to process the operating characteristics and the operating condition information using a trained estimation model to obtain an estimated remaining life, Wherein, the estimation model is a machine learning model, and the training samples of the estimation model include full life cycle operation signals of the gearbox under multiple different working conditions.

8. An electronic device, characterized in that: include: at least one processor; at least one memory storing computer executable instructions, Wherein, when the computer executable instructions are executed by the at least one processor, the at least one processor is prompted to execute the gearbox life prediction method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: When the instructions in the computer-readable storage medium are executed by at least one processor, the instructions cause the at least one processor to perform the gearbox life prediction method according to any one of claims 1 to 6.

10. A computer program product comprising computer instructions, characterized in that: When the computer instructions are executed by at least one processor, the at least one processor is prompted to perform the gearbox life prediction method according to any one of claims 1 to 6.

Images