Bearing life prediction method, system and device based on neural network

Abstract

The present invention discloses a method, system and device for predicting bearing life based on a neural network. The method includes: obtaining historical relevant data of the bearing during operation for preprocessing, extracting data features and sorting them, and obtaining a relevant feature matrix; processing the preprocessed bearing relevant data based on a bearing rated life model to obtain a bearing rated life data set; constructing a bearing life prediction pretraining model and training it to obtain a bearing life prediction model; predicting the next time period relevant data based on the current time period relevant data based on the bearing life prediction model to obtain bearing life prediction parameters; inputting the bearing life prediction parameters into a preset prediction bearing actual life model to obtain a predicted bearing actual life. The method solves the problem that the existing bearing actual working life calculation method does not fully consider the actual working conditions, resulting in low calculation accuracy, and can effectively predict the actual working life of the bearing with high accuracy and reliability.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a bearing life prediction method, system and device based on a neural network. Background Art

[0002] Rolling bearings are precision mechanical components used to support rotating shafts and parts on the shafts, maintaining the normal working position and rotation accuracy of the shafts. The motors of rotating equipment such as gearboxes, water pumps, and fans all use a large number of rolling bearings to achieve the corresponding functions. Due to improper lubrication, excessive loads, leakage current and other factors, the performance of rolling bearings may deteriorate or even fail, which may cause the entire mechanical system to fail during operation. Therefore, it is important to manage the health of rolling bearings throughout their life cycle during the operation of the mechanical system.

[0003] In the prior art, the rated life calculation formula of rolling bearings is usually used to calculate the rated life of rolling bearings, but this method cannot take into account the actual working state and working conditions of rolling bearings, resulting in inaccurate calculation results. In addition, PHM (Prognostics and Health Management) bearing fault prediction and health management technology is also used to manage the entire life cycle of rolling bearings. This method mainly relies on traditional signal processing and statistical methods, so the working process and calculation process are very complicated, and the lack or change of certain data may lead to inaccurate results, which in turn leads to management loopholes.

[0004] It can be seen that the current life cycle health management method will lead to inaccurate results. How to make the calculation results more accurate and the prediction accuracy higher is a hot topic in current research based on artificial intelligence technology. Therefore, how to combine the health management of the entire life cycle of rolling bearings through artificial intelligence technology and improve the accuracy of warning and prediction is the problem to be solved by the present invention. Summary of the invention

[0005] In view of the deficiencies in the prior art, the present invention discloses a bearing life prediction method, system and device based on a neural network.

[0006] In order to achieve the above technical effects, the present invention adopts the following technical solutions:

[0007] A bearing life prediction method based on a neural network comprises the following steps:

[0008] Obtain historical data related to the working of the bearing, preprocess the historical data to obtain preprocessed bearing related data, extract data features and sort them to obtain a related feature matrix, and build a preset prediction model for the actual life of the bearing based on the related feature matrix;

[0009] Processing the pre-processed bearing-related data based on a basic bearing rating life model to obtain a basic bearing rating life data set;

[0010] Constructing a bearing life prediction pre-training model, and training the bearing life prediction pre-training model according to the bearing rated life data set to obtain a bearing life prediction model;

[0011] Inputting the relevant data of the current period into the bearing life prediction model to predict the relevant data of the next period to obtain the bearing life prediction parameters;

[0012] The bearing life prediction parameters are processed by a preset model for predicting the actual bearing life to obtain the predicted actual bearing life.

[0013] As an implementable embodiment, the historical related data includes one or more of vibration signal data, motor current signal data, temperature signal data, bearing type data, bearing basic rated dynamic load data, bearing equivalent dynamic load data and bearing actual working life data.

[0014] As an implementable embodiment, the basic bearing rating life model is expressed as follows:

[0015]

[0016] in, represents the basic rated life, Indicates the basic dynamic load rating, is the equivalent dynamic load on the bearing, An index indicating the life formula for different bearing types.

[0017] As an implementable method, the construction of a bearing life prediction pre-training model includes the following steps:

[0018] The bearing life prediction pre-training model is constructed based on the improved BP neural network model of multi-layer perceptron. The input layer receives the bearing rated life data set, and the output layer outputs the bearing life prediction parameters. The nonlinear result is obtained through the activation function.

[0019] Construct a loss function and evaluate and adjust the model parameters of the bearing life prediction pre-training model through back propagation of the loss function;

[0020] An improved AFSA algorithm model was constructed, and the bearing life prediction pre-training model was optimized by the improved AFSA algorithm model to obtain the optimal model hyperparameters.

[0021] As an implementable method, the construction of the improved AFSA algorithm model includes the following steps:

[0022] Determine the dynamic field of view and dynamic step size of the artificial fish school, as follows:

[0023]

[0024]

[0025] The current fish queries the fish with the highest food concentration based on the dynamic field of view and determines the position of the fish with the highest food concentration, wherein the position of the fish with the highest food concentration includes m The data features related to bearing operation and life are used to calculate the attraction of the fish with the highest food concentration in the field of view to the current fish, which is expressed as follows:

[0026]

[0027] in, Represents dynamic field of view, represents the dynamic step size, represents the maximum dynamic field of view, represents the maximum dynamic step size, Indicates attraction, It means to avoid positive numbers with denominator of 0. Indicates the location of the fish with the highest food concentration, Indicates the food concentration at the fish location with the highest food concentration, , Indicates The sample Data features, Indicates the current fish position. represents the number of iterations, represents the field of view adjustment coefficient, Represents the step size adjustment coefficient;

[0028] According to the average position of the fish school, the current fish position and the position of the fish with the highest food concentration, the current fish movement direction is determined, which is expressed as follows:

[0029]

[0030] in, Indicates the direction of movement, They represent weight coefficients respectively, represents the average position of the fish school, , Indicates the number of fish. Indicates Fish position;

[0031] According to the current fish position, the current fish moving direction and the current fish dynamic step length, the current fish moving position is obtained, which is expressed as follows:

[0032]

[0033] in, Indicates the current position of the fish after moving;

[0034] The school of fish conducts random exploration with a preset probability, selects a random direction and each component obeys a uniform distribution, updates the position of the current fish after movement according to the current fish position, and obtains the updated position of the current fish after movement, which is expressed as follows:

[0035]

[0036] in, Indicates the position of the current fish after the update. Represents a random direction.

[0037] As an implementable method, the following steps are also included:

[0038] The accuracy of the bearing life prediction model is judged based on the error model to obtain a judgment result. The error model is expressed as follows:

[0039]

[0040] in, Indicates the actual life value, represents the life expectancy prediction value, Indicates the number of times, Indicates the error result.

[0041] As an implementable method, the activation function is expressed as follows:

[0042]

[0043] The first-order derivative of the activation function is expressed as follows:

[0044]

[0045] in, Indicates input, Represents coefficients or weights.

[0046] As an implementable method, the loss function is expressed as follows:

[0047]

[0048] in, Indicates the bearing rating life, represents the predicted actual bearing life, represents the parameters of the loss function, W Represents the current model parameter set, represents the optimal set of model parameters.

[0049] As an implementable method, the preset model for predicting the actual life of the bearing is expressed as follows:

[0050]

[0051] in, It indicates the predicted actual bearing life. and They are respectively represented as bearing life prediction compensation parameters, They represent the relevant feature matrices respectively.

[0052] A bearing life prediction system based on a neural network is implemented based on a general control module and includes:

[0053] The working condition acquisition module is used to obtain the historical relevant data of the bearing when it is working, preprocess the historical relevant data to obtain the preprocessed bearing relevant data, extract the data features and sort them to obtain the relevant feature matrix;

[0054] A data processing module processes the pre-processed bearing-related data based on a bearing rated life model to obtain a bearing rated life data set;

[0055] A model building module is used to build a bearing life prediction pre-training model, and train the bearing life prediction pre-training model according to the bearing rated life data set to obtain a bearing life prediction model;

[0056] A prediction and reasoning module, which inputs the relevant data of the current period into the bearing life prediction model to predict the relevant data of the next period and obtain the bearing life prediction parameters;

[0057] The result calculation module is used to process the bearing life prediction parameters through a preset prediction model for the actual bearing life to obtain the predicted actual bearing life, wherein the preset prediction model for the actual bearing life is constructed based on the relevant feature matrix.

[0058] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the following method is implemented:

[0059] Obtain historical data related to the working of the bearing, preprocess the historical data to obtain preprocessed bearing related data, extract data features and sort them to obtain a related feature matrix, and build a preset prediction model for the actual life of the bearing based on the related feature matrix;

[0060] Processing the pre-processed bearing-related data based on a basic bearing rating life model to obtain a basic bearing rating life data set;

[0061] Constructing a bearing life prediction pre-training model, and training the bearing life prediction pre-training model according to the bearing rated life data set to obtain a bearing life prediction model;

[0062] Inputting the relevant data of the current period into the bearing life prediction model to predict the relevant data of the next period to obtain the bearing life prediction parameters;

[0063] The bearing life prediction parameters are processed by a preset model for predicting the actual bearing life to obtain the predicted actual bearing life.

[0064] A bearing life prediction device based on a neural network comprises a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor implements the following method when executing the computer program:

[0065] Obtain historical data related to the working of the bearing, preprocess the historical data to obtain preprocessed bearing related data, extract data features and sort them to obtain a related feature matrix, and build a preset prediction model for the actual life of the bearing based on the related feature matrix;

[0066] Processing the pre-processed bearing-related data based on a basic bearing rating life model to obtain a basic bearing rating life data set;

[0067] Constructing a bearing life prediction pre-training model, and training the bearing life prediction pre-training model according to the bearing rated life data set to obtain a bearing life prediction model;

[0068] Inputting the relevant data of the current period into the bearing life prediction model to predict the relevant data of the next period to obtain the bearing life prediction parameters;

[0069] The bearing life prediction parameters are processed by a preset model for predicting the actual bearing life to obtain the predicted actual bearing life.

[0070] The present invention has significant technical effects due to the adoption of the above technical solution:

[0071] The present invention provides a bearing life prediction method based on a neural network; based on various data collected when the bearing is working, including vibration data, motor current data, temperature data, bearing type, basic rated dynamic load of the bearing, equivalent dynamic load of the bearing, and actual working life of the bearing, data preprocessing and feature extraction are performed to establish a bearing life prediction model based on a neural network, and the neural network is trained using an improved AFSA algorithm to optimize the hyperparameter combination; the bearing life is predicted according to the bearing life prediction model; on the basis of the bearing rated life model, the credibility and stability of the algorithm are improved, and the problem that the existing calculation method for the actual working life of the bearing does not fully consider the actual working conditions, resulting in low calculation accuracy, can effectively predict the actual working life of the bearing, and has high accuracy and reliability. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0073] Figure 1 It is a schematic flow diagram of the method of the present invention;

[0074] Figure 2 It is a schematic diagram of the structure of the system of the present invention. DETAILED DESCRIPTION

[0075] The technical solutions in the embodiments of the present invention are described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0076] Embodiment 1:

[0077] A bearing life prediction method based on neural network, such as Figure 1 As shown, the following steps are included:

[0078] S100, obtaining historical relevant data of the bearing during operation, preprocessing the historical relevant data to obtain preprocessed bearing relevant data, extracting and sorting data features to obtain a relevant feature matrix, and constructing a preset prediction model for the actual life of the bearing based on the relevant feature matrix;

[0079] S200, processing the pre-processed bearing-related data based on a basic bearing rated life model to obtain a basic bearing rated life data set;

[0080] S300, constructing a bearing life prediction pre-training model, and training the bearing life prediction pre-training model according to the bearing rated life data set to obtain a bearing life prediction model;

[0081] S400, inputting the relevant data of the current period into the bearing life prediction model to predict the relevant data of the next period, and obtaining the bearing life prediction parameters;

[0082] S500: Process the bearing life prediction parameters by using a preset bearing actual life prediction model to obtain a predicted bearing actual life.

[0083] In one embodiment, the bearing is a roller bearing, and a high-precision sensor is used to collect historical relevant data of the roller bearing during operation, including vibration signal data, motor current signal data, temperature signal data, bearing type data, bearing basic rated dynamic load data, bearing equivalent dynamic load data and bearing actual working life data, and ensure the accuracy and completeness of data collection to provide a reliable data basis for subsequent analysis; the collected roller bearing data is preprocessed, and a data filtering algorithm is used to remove abnormal data and noise interference when the roller bearing is working to improve the data quality; low-pass filtering, high-pass filtering and band-pass filtering are used to effectively remove noise and interference signals in different frequency bands; Z-score standardization is used to normalize the data to make the data comparable and stable; downsampling technology is used to reduce the amount of data while retaining the key features of the data; frequency domain transformation is performed according to fast Fourier transform to extract frequency domain features; feature extraction is performed on the preprocessed roller bearing signal data.

[0084] According to the signal processing algorithm, features are extracted from the vibration signal data, and the features include vibration mean, vibration root mean square value, vibration peak value, vibration variance, vibration kurtosis, vibration skewness, spectrum main peak frequency, first harmonic frequency, first harmonic amplitude, and second harmonic frequency; according to the current waveform, features of the motor current signal data are extracted to obtain the influence of different working conditions of the motor on the roller bearing, and the features include current mean, current effective value, current peak value, current harmonic content, current spectrum main peak frequency, current spectrum sub-peak frequency, current spectrum peak ratio, and current fluctuation amplitude; according to the temperature measurement technology, features are extracted from the temperature signal data to obtain the thermal state of the roller bearing when working, and the features include temperature mean, temperature maximum value, temperature minimum value, temperature change rate, and temperature standard deviation.

[0085] After extracting various features, they will be analyzed to determine the impact of each feature data on the bearing life, and the relevant feature matrix will be obtained by sorting them from large to small according to the impact.

[0086] Substitute the pre-processed roller bearing type data, bearing basic rated dynamic load data, bearing equivalent dynamic load data, bearing actual working life data, and corresponding life reliability adjustment coefficient and life correction coefficient into the bearing rated life model to calculate the rated life of SKF bearings and obtain the data set of roller bearing rated life. The basic bearing rated life model used here is ,in, represents the basic rated life, Indicates the basic dynamic load rating, is the equivalent dynamic load on the bearing, The life formula index of different bearing types is shown in FIG. The value is 10 / 3.

[0087] In step S300, a bearing life prediction pre-training model is constructed, and the bearing life prediction pre-training model is trained according to the bearing rated life data set to obtain a bearing life prediction model, which specifically includes the following steps:

[0088] The bearing life prediction pre-training model is constructed based on the improved BP neural network model of multi-layer perceptron. The input layer receives the bearing rated life data set, and the output layer outputs the bearing life prediction parameters. The nonlinear result is obtained through the activation function.

[0089] Construct a loss function and evaluate and adjust the model parameters of the bearing life prediction pre-training model through back propagation of the loss function;

[0090] An improved AFSA algorithm model was constructed, and the bearing life prediction pre-training model was optimized by the improved AFSA algorithm model to obtain the optimal model hyperparameters.

[0091] BP neural network, whose full name is back propagation neural network, is a multi-layer feedforward neural network that trains network weights through a back propagation algorithm to achieve the purpose of classifying or regressing input data. BP neural network is one of the very basic and important models in the field of deep learning. The BP neural network architecture adopted in this embodiment, in combination with practical applications, constructs a loss function and improves the AFSA algorithm model, thereby avoiding overfitting or reducing the training time when training the bearing life prediction pre-training model.

[0092] The optimization is achieved by simulating the foraging behavior of fish schools based on the characteristic that the place with the largest number of fish in the water area is the place with the most nutrients in the water area. It mainly uses the four basic behaviors of fish: random, foraging, clustering and tail-chasing behavior. It has the advantage of fast convergence speed and can be used to solve problems with high real-time requirements. Artificial fish is an abstract and virtualized entity of real fish. It can receive environmental stimulus information and perform corresponding activities. Its state at the next moment depends on its own state and the state of the environment. It affects the environment through its own activities, and then affects the activities of other artificial fish. The construction of the improved AFSA algorithm model includes the following steps:

[0093] The construction of the improved AFSA algorithm model includes the following steps:

[0094] Determine the dynamic field of view and dynamic step size of the artificial fish school, as follows:

[0095]

[0096]

[0097] The current fish queries the fish with the highest food concentration based on the dynamic field of view and determines the position of the fish with the highest food concentration, wherein the position of the fish with the highest food concentration includes m The data features related to bearing operation and life are used to calculate the attraction of the fish with the highest food concentration in the field of view to the current fish, which is expressed as follows:

[0098]

[0099] in, Represents dynamic field of view, represents the dynamic step size, represents the maximum dynamic field of view, represents the maximum dynamic step size, Indicates attraction, It means to avoid positive numbers with denominator of 0. Indicates the location of the fish with the highest food concentration, Indicates the food concentration at the fish location with the highest food concentration, , Indicates The sample Data features, Indicates the current fish position. represents the number of iterations, represents the field of view adjustment coefficient, represents the step adjustment coefficient. Here, the data characteristics include: characteristics of vibration signals (such as vibration root mean square value, vibration peak value, etc.), characteristics of motor current signals (such as current mean value, current peak value, etc.), characteristics of temperature signals (such as temperature change rate, maximum temperature, minimum temperature, etc.);

[0100] According to the average position of the fish school, the current fish position and the position of the fish with the highest food concentration, the current fish movement direction is determined, which is expressed as follows:

[0101]

[0102] in, Indicates the direction of movement, They represent weight coefficients respectively, represents the average position of the fish school, , Indicates the number of fish. Indicates Fish position;

[0103] According to the current fish position, the current fish moving direction and the current fish dynamic step length, the current fish moving position is obtained, which is expressed as follows:

[0104]

[0105] in, Indicates the current position of the fish after moving;

[0106] The school of fish conducts random exploration with a preset probability, selects a random direction and each component obeys a uniform distribution, updates the position of the current fish after movement according to the current fish position, and obtains the updated position of the current fish after movement, which is expressed as follows:

[0107]

[0108] in, Indicates the position of the current fish after the update. Represents a random direction.

[0109] The optimization target of the improved artificial fish swarm algorithm is the neural network. The parameters to be determined are the number of hidden layers, the number of neurons in each layer, the learning rate, and the regularization parameter L. 1 , L 2 , loss function parameter α, and activation function parameter β.

[0110] In a specific embodiment, the following steps are also included:

[0111] The accuracy of the bearing life prediction model is judged based on the error model to obtain a judgment result. The error model is expressed as follows:

[0112]

[0113] in, Indicates the actual life value, represents the life expectancy prediction value, Indicates the number of times, Indicates the error result.

[0114] Activation functions are able to transform linear combinations of inputs into nonlinear outputs, which is crucial for solving complex nonlinear problems. Without nonlinear activation functions, no matter how many layers a neural network has, it can only learn linear combinations of inputs, which limits the network's expressive power. During training, activation functions affect the propagation of gradients. A suitable activation function can help the gradient flow effectively in the network, avoiding the problem of gradient vanishing or gradient exploding, thereby helping the network converge. By using different activation functions, neural networks can learn more complex function mappings. For example, the ReLU (Rectified Linear Unit) activation function is widely used because of its high computational efficiency and good gradient propagation. Different activation functions have different output ranges. For example, the Sigmoid function limits the output to between (0,1), while the Tanh function has an output range of (-1,1). This can affect the design of the network's output layer, especially when performing classification tasks. Some activation functions such as Sigmoid and Tanh have gradients close to zero when the input values ​​are large or small, which can cause the network weights to update slowly. The ReLU activation function has a constant gradient in the positive interval, which helps the network learn faster and can increase the sparsity of the network, that is, there are more zero values ​​in the output of the network. In this embodiment, the activation function improved based on the ReLU activation function is expressed as follows:

[0115]

[0116] This activation function can enhance the expressiveness of the model by dynamically adjusting the nonlinear transformation of the input while maintaining the complexity of the model. Compared with other activation functions, this function can more effectively deal with potential outliers and diverse input features in bearing life prediction while taking into account gradient stability, so it shows higher accuracy and robustness in our prediction task. This choice makes our model not only strike a balance between complexity and computational efficiency, but also has stronger adaptability in different application scenarios.

[0117] The first-order derivative of the activation function is expressed as follows:

[0118]

[0119] in, Indicates input, Represents coefficients or weights.

[0120] In addition, a loss function is also designed. The loss function in this embodiment is:

[0121]

[0122] in, Indicates the bearing rating life, represents the predicted actual bearing life, represents the parameters of the loss function, W Represents the current set of model parameters (all weight matrices and bias vectors in the neural network), which are continuously updated during the training process to minimize the value of the loss function. This is the optimal parameter combination.

[0123] In addition, in one embodiment, the preset model for predicting the actual life of the bearing is expressed as follows:

[0124]

[0125] in, and They are respectively represented as bearing life prediction compensation parameters, Respectively represent the relevant feature matrix. The bearing life prediction model is used to predict the relevant data of the next period, and the bearing life prediction parameters are input into the preset prediction bearing actual life model to finally obtain the predicted bearing actual life. .

[0126] Embodiment 2:

[0127] A bearing life prediction system based on neural network is implemented based on the overall control module, such as Figure 2 As shown, including:

[0128] The working condition acquisition module 100 is used to acquire the historical relevant data of the bearing when it is working, pre-process the historical relevant data to obtain the pre-processed bearing relevant data, extract the data features and sort them to obtain the relevant feature matrix;

[0129] A data processing module 200 processes the pre-processed bearing-related data based on a bearing rated life model to obtain a bearing rated life data set;

[0130] A model building module 300 is used to build a bearing life prediction pre-training model, and train the bearing life prediction pre-training model according to the bearing rated life data set to obtain a bearing life prediction model;

[0131] The prediction and reasoning module 400 inputs the current period related data into the bearing life prediction model to predict the next period related data and obtain the bearing life prediction parameters;

[0132] The result calculation module 500 is used to process the bearing life prediction parameters through a preset prediction model for the actual bearing life to obtain the predicted actual bearing life, wherein the preset prediction model for the actual bearing life is constructed based on a related feature matrix.

[0133] Various changes and modifications can be made without departing from the spirit and scope of the present invention, and all equivalent technical solutions also belong to the scope of the present invention.

[0134] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0135] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, devices or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0136] The present invention is described with reference to the flowcharts and / or block diagrams of the method, terminal device (system), and computer program product according to the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing terminal device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0137] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing terminal device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0138] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device so that a series of operating steps are executed on the computer or other programmable terminal device to produce a computer-implemented process, thereby providing instructions for executing on the computer or other programmable terminal device to implement the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0139] It should be noted that:

[0140] The "one embodiment" or "embodiment" mentioned in the specification means that the specific features, structures or characteristics described in conjunction with the embodiment are included in at least one embodiment of the present invention. Therefore, the phrases "one embodiment" or "embodiment" appearing in various places throughout the specification do not necessarily refer to the same embodiment.

[0141] In addition, it should be noted that the shapes and names of the parts and components of the specific embodiments described in this specification may be different. Any equivalent or simple changes made based on the structure, features and principles described in the patent concept of the present invention are included in the protection scope of the patent of the present invention. The technicians in the technical field of the present invention can make various modifications or supplements to the specific embodiments described or replace them in a similar manner, as long as they do not deviate from the structure of the present invention or exceed the scope defined by the claims, they should all fall within the protection scope of the present invention.

Claims

1. A bearing life prediction method based on neural network, characterized in that: The following steps are involved: Obtain historical data related to the working of the bearing, preprocess the historical data to obtain preprocessed bearing related data, extract data features and sort them to obtain a related feature matrix, and build a preset prediction model for the actual bearing life based on the related feature matrix and bearing life prediction parameters; Processing the pre-processed bearing-related data based on a basic bearing rating life model to obtain a basic bearing rating life data set; Constructing a bearing life prediction pre-training model, and training the bearing life prediction pre-training model according to the basic bearing rated life data set to obtain a bearing life prediction model; Inputting the relevant data of the current period into the bearing life prediction model to predict the relevant data of the next period to obtain the bearing life prediction parameters; The bearing life prediction parameters are processed by a preset bearing actual life prediction model to obtain a predicted bearing actual life; The historical related data includes one or more of vibration signal data, motor current signal data, temperature signal data, bearing type data, bearing basic rated dynamic load data, bearing equivalent dynamic load data and bearing actual working life data; The basic bearing rating life model is expressed as follows: in, represents the basic rated life, Indicates the basic dynamic load rating, is the equivalent dynamic load on the bearing, Indices representing the life formulas of different bearing types; The construction of a bearing life prediction pre-training model comprises the following steps: The bearing life prediction pre-training model is constructed based on the improved BP neural network model of multi-layer perceptron. The input layer receives the bearing rated life data set, and the output layer outputs the bearing life prediction parameters. The nonlinear result is obtained through the activation function. Construct a loss function and evaluate and adjust the model parameters of the bearing life prediction pre-training model through back propagation of the loss function; An improved AFSA algorithm model was constructed, and the bearing life prediction pre-training model was optimized by the improved AFSA algorithm model to obtain the optimal model hyperparameters; Wherein, the construction of the improved AFSA algorithm model comprises the following steps: Determine the dynamic field of view and dynamic step size of the artificial fish school, as follows: The current fish queries the fish with the highest food concentration based on the dynamic field of view and determines the position of the fish with the highest food concentration, wherein the position of the fish with the highest food concentration includes m The data features related to bearing operation and life are used to calculate the attraction of the fish with the highest food concentration in the field of view to the current fish, which is expressed as follows: in, Represents dynamic field of view, represents the dynamic step size, represents the maximum dynamic field of view, represents the maximum dynamic step size, Indicates attraction, It means to avoid positive numbers with denominator of 0. Indicates the location of the fish with the highest food concentration, Indicates the food concentration at the fish location with the highest food concentration, , Indicates The sample Data features, Indicates the current fish position. represents the number of iterations, represents the field of view adjustment coefficient, Represents the step size adjustment coefficient; According to the average position of the fish school, the current fish position and the position of the fish with the highest food concentration, the current fish movement direction is determined, which is expressed as follows: in, Indicates the direction of movement, They represent weight coefficients respectively, represents the average position of the fish school, , Indicates the number of fish, Indicates Fish position; According to the current fish position, the current fish moving direction and the current fish dynamic step length, the current fish moving position is obtained, which is expressed as follows: in, Indicates the current position of the fish after moving; The school of fish conducts random exploration with a preset probability, selects a random direction and each component obeys a uniform distribution, updates the position of the current fish after movement according to the current fish position, and obtains the updated position of the current fish after movement, which is expressed as follows: in, Indicates the position of the current fish after the update. represents random direction; The preset model for predicting the actual life of the bearing is expressed as follows: in, It indicates the predicted actual bearing life. and They are respectively expressed as bearing life prediction parameters, Respectively represent the relevant feature matrix; Wherein, the loss function is expressed as follows: in, Indicates the bearing rating life, represents the predicted actual bearing life, represents the parameters of the loss function, W Represents the current model parameter set, represents the optimal set of model parameters.

2. A method for predicting bearing life based on a neural network according to claim 1, characterized in that: The following steps are also included: The accuracy of the bearing life prediction model is judged based on the error model to obtain a judgment result. The error model is expressed as follows: in, Indicates the actual life value, represents the life expectancy prediction value, Indicates the number of times, Indicates the error result.

3. The method for predicting bearing life based on a neural network according to claim 1, characterized in that: The activation function is expressed as follows: The first-order derivative of the activation function is expressed as follows: in, Indicates input, Represents coefficients or weights.

4. A bearing life prediction system based on a neural network, implemented based on a general control module, characterized in that: include: The working condition acquisition module is used to obtain the historical data related to the bearing during operation, preprocess the historical data related to obtain the preprocessed bearing related data, extract the data features and sort them to obtain the relevant feature matrix, and build a preset prediction model for the actual bearing life based on the relevant feature matrix and the bearing life prediction parameters; A data processing module processes the pre-processed bearing-related data based on a basic bearing rated life model to obtain a basic bearing rated life data set; A model building module is used to build a bearing life prediction pre-training model, and train the bearing life prediction pre-training model according to the basic bearing rated life data set to obtain a bearing life prediction model; A prediction and reasoning module, which inputs the relevant data of the current period into the bearing life prediction model to predict the relevant data of the next period and obtain the bearing life prediction parameters; A result calculation module is used to process the bearing life prediction parameters through a preset prediction model for the actual bearing life to obtain a predicted actual bearing life; Wherein, the historical related data includes one or more of vibration signal data, motor current signal data, temperature signal data, bearing type data, bearing basic rated dynamic load data, bearing equivalent dynamic load data and bearing actual working life data; The basic bearing rating life model is expressed as follows: in, represents the basic rated life, Indicates the basic dynamic load rating, is the equivalent dynamic load on the bearing, Indices representing the life formulas of different bearing types; The construction of a bearing life prediction pre-training model comprises the following steps: The bearing life prediction pre-training model is constructed based on the improved BP neural network model of multi-layer perceptron. The input layer receives the bearing rated life data set, and the output layer outputs the bearing life prediction parameters. The nonlinear result is obtained through the activation function. Construct a loss function and evaluate and adjust the model parameters of the bearing life prediction pre-training model through back propagation of the loss function; An improved AFSA algorithm model was constructed, and the bearing life prediction pre-training model was optimized by the improved AFSA algorithm model to obtain the optimal model hyperparameters; Wherein, the construction of the improved AFSA algorithm model comprises the following steps: Determine the dynamic field of view and dynamic step size of the artificial fish school, as follows: The current fish queries the fish with the highest food concentration based on the dynamic field of view and determines the position of the fish with the highest food concentration, wherein the position of the fish with the highest food concentration includes m The data features related to bearing operation and life are used to calculate the attraction of the fish with the highest food concentration in the field of view to the current fish, which is expressed as follows: in, Represents dynamic field of view, represents the dynamic step size, represents the maximum dynamic field of view, represents the maximum dynamic step size, Indicates attraction, It means to avoid positive numbers with denominator of 0. Indicates the location of the fish with the highest food concentration, Indicates the food concentration at the fish location with the highest food concentration, , Indicates The sample Data features, Indicates the current fish position. represents the number of iterations, represents the field of view adjustment coefficient, Represents the step size adjustment coefficient; According to the average position of the fish school, the current fish position and the position of the fish with the highest food concentration, the current fish movement direction is determined, which is expressed as follows: in, Indicates the direction of movement, They represent weight coefficients respectively, represents the average position of the fish school, , Indicates the number of fish, Indicates Fish position; According to the current fish position, the current fish moving direction and the current fish dynamic step length, the current fish moving position is obtained, which is expressed as follows: in, Indicates the current position of the fish after moving; The school of fish conducts random exploration with a preset probability, selects a random direction and each component obeys a uniform distribution, updates the position of the current fish after movement according to the current fish position, and obtains the updated position of the current fish after movement, which is expressed as follows: in, Indicates the position of the current fish after the update. represents random direction; The preset model for predicting the actual life of the bearing is expressed as follows: in, It indicates the predicted actual bearing life. and They are respectively expressed as bearing life prediction parameters, Respectively represent the relevant feature matrix; Wherein, the loss function is expressed as follows: in, Indicates the bearing rating life, represents the predicted actual bearing life, represents the parameters of the loss function, W Represents the current model parameter set, represents the optimal set of model parameters.

5. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 3 is implemented.

6. A bearing life prediction device based on a neural network, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 3 is implemented.

Images