A mechanical bearing life prediction method based on exponential smoothing and bidirectional long short-term memory network
By combining exponential smoothing and bidirectional long and short-term memory networks, the accuracy of mechanical bearing life prediction in the prior art is solved, the learning ability of the model is improved, the accurate prediction of mechanical bearing life is achieved, and the maintenance cost is reduced.
Patent Information
- Application Number
- CN202111450971.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-12-01
- Publication Date
- 2024-12-31
- Estimated Expiration
- 2041-12-01
AI Technical Summary
The prior art is difficult to accurately identify mechanical bearings that exceed their service life, resulting in mechanical equipment failures, and the Bi-LSTM model only learns past states, ignoring the impact of future states, resulting in limited data learning capabilities.
Using an exponential smoothing and bidirectional long and short-term memory network, the model is trained to predict the life of mechanical bearings by obtaining the historical operation and maintenance data of the equipment, and using the Bi-LSTM network combined with exponential smoothing processing and feature enhancement.
It improves the model's learning ability on the front and rear dependencies, can more accurately predict the life of mechanical bearings, reduces maintenance costs, and ensures the normal operation of the equipment.
Smart Images
Figure CN114091351B_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to mechanical bearing life prediction, and specifically is a mechanical bearing life prediction method based on exponential smoothing and bidirectional long short-term memory network. Background Art
[0002] Mechanical bearings are key components in the transmission system of mechanical equipment. Whether the bearings are in normal operation or not plays an important role in ensuring the safe operation of mechanical equipment. However, in actual production operations, some bearings that have exceeded their service life are often still running in mechanical equipment. If they are not replaced in time, large-scale mechanical failures will occur. If bearings that have exceeded their service life can be accurately identified and replaced, maintenance funds can be effectively saved, the normal operation of mechanical equipment can be guaranteed to the greatest extent, and the production and operation of the enterprise can be protected.
[0003] The Bi-LSTM (Bidirectional Long Short-Term Memory Network) neural network structure model is divided into two independent LSTMs. The input sequence is input into the two LSTM neural networks in forward and reverse order for feature extraction, and the output result is also the splicing of the results in the two directions. The design concept of the Bi-LSTM (Bidirectional Long Short-Term Memory Network) model is to make the feature data obtained at time t have information between the past and the future at the same time. Experiments have shown that this neural network structure model has better efficiency and performance in text feature extraction than a single LSTM structure model. LSTM only learns the impact of the past state on the current state, ignores the role of the future state, does not fully utilize the before-after dependency relationship of the time series, and has limited learning ability for data. The Bi-LSTM network (Bidirectional Long Short-Term Memory Network) can simultaneously learn the influence of the past and future states on the current state, greatly improving the model's learning ability for long-sequence data with before-after dependencies. The Bi-LSTM network (Bidirectional Long Short-Term Memory Network) can independently process the time data through the forward layer and the backward layer, and feed the processing results of the two layers forward to the output layer at the same time, which can fully utilize the life information contained in the data of the past and future moments.
[0004] Exponential smoothing is actually a special weighted moving average method. Its characteristics are: First, the exponential smoothing method further strengthens the role of recent observations in the observation period on the predicted value, and the weights assigned to observations at different times are different, thereby increasing the weights of recent observations, so that the predicted value can quickly reflect the actual changes in the market. The weights are reduced in a geometric series, and the first term of this series is the smoothing constant a, and the common ratio is (1-a). Second, the exponential smoothing method is flexible in the weights assigned to the observations, and different a values can be taken to change the rate of change of the weights. If a takes a small value, the weight changes more quickly, and the recent change trend of the observations can be more quickly reflected in the exponential moving average. Therefore, using the exponential smoothing method, different a values can be selected to adjust the uniformity of the time series observations (that is, the stability of the trend change). Summary of the invention
[0005] In order to solve the above technical problems, the purpose of the present invention is to provide a mechanical bearing life prediction method based on exponential smoothing and bidirectional long short-term memory network.
[0006] To achieve the above object, the present invention adopts the following technical solution:
[0007] A mechanical bearing life prediction method based on exponential smoothing and bidirectional long short-term memory network, comprising:
[0008] Obtain historical operation and maintenance data of the equipment, and organize and archive the equipment data in the form of pre-processed data;
[0009] Performing data cleaning and feature preprocessing on the device data to obtain processed data;
[0010] The processed data is smoothed and then input into a trained Bi-LSTM (bidirectional long short-term memory network) to obtain an output result, which is the mechanical bearing life prediction result.
[0011] It also includes training methods for exponential smoothing and Bi-LSTM networks (bidirectional long short-term memory networks):
[0012] The acquired equipment historical operation and maintenance data are subjected to exponential smoothing, and then a time window is introduced to enhance the features;
[0013] Build a Bi-LSTM network (bidirectional long short-term memory network) structure, which includes several Bi-LSTM (bidirectional long short-term memory networks) connected in sequence and N fully connected layers;
[0014] Divide the sample data into training set and test set;
[0015] Set up the training plan: how many times to loop over the training set, the number of training batches, random shuffling, optimizer, loss function, and saving conditions;
[0016] Inputting the training set into the Bi-LSTM network (bidirectional long short-term memory network), and training according to the training scheme to obtain a preliminary training model;
[0017] The exponential smoothing process and each Bi-LSTM (Bidirectional Long Short-Term Memory Network) block includes sequentially connected units, BN layers and ReLU activation functions.
[0018] The exponential smoothing process uses a weighted observation value calculation method proposed in the Python data analysis library Pandas, that is, each coefficient is a general weighted average of the exponential function: For any given α, in calculating y t When x is far away from the current time t-i The weight is small and its influence on the calculation can be ignored, so it can be understood as x within a certain range s at the near end. t-i Plays a major role, namely {x t-i |0≤i≤s-1}, s is regarded as the smoothing coefficient, and the relationship between α and s is as follows:
[0019] The mathematical transformation formula of the Bi-LSTM (Bidirectional Long Short-Term Memory Network) block includes:
[0020] h t =f(w1x t+ wxya t-1 )
[0021] h′ t =f(w3x t +w5h′ t+1 )
[0022] O t =g(w4h t +w6h′ t )
[0023] Where: h t , h′ t are the outputs of the forward and backward propagation layers and the backward propagation layer at time t; w1 and w3 are the weight matrices of the input layer and the backward propagation layer respectively; w3 and w5 are the weight matrices from the forward and backward propagation layers to their own propagation layers respectively; w4 and w6 are the weight matrices from the forward and backward propagation layers to the output layer respectively; t is the output value of the final output gate; function f(·) represents the cell calculation process; function g(·) represents the function for concatenating the forward and backward propagation results.
[0024] The calculation of the BN layer includes:
[0025]
[0026]
[0027] Among them, x p is the input of the pth batch, E[x p ] is x p The average value of Var[x p ] is x p The standard deviation of , γ and β are adaptively updated learnable parameters, y p is the output of the BN layer neurons.
[0028] The calculation of the ReLU activation function includes:
[0029] Where x is the function input.
[0030] Beneficial effects: The mechanical bearing life prediction method based on exponential smoothing and bidirectional long short-term memory network of the present invention is based on historical operating data and does not need to establish a Bi-LSTM (bidirectional long short-term memory network) network model of a physical loss model. It is based on a large amount of historical operating data for prediction, and its applicability and scalability will be improved. BRIEF DESCRIPTION OF THE DRAWINGS
[0031] Figure 1 The present invention is a flow chart of mechanical bearing life prediction based on exponential smoothing and bidirectional long short-term memory network.
[0032] Figure 2 This is a composition diagram of a bidirectional long short-term memory network.
[0033] Figure 3 Bearing life prediction model based on bidirectional long short-term memory network. DETAILED DESCRIPTION
[0034] The present invention is described in detail below in conjunction with the accompanying drawings.
[0035] The present invention discloses a mechanical bearing life prediction method based on exponential smoothing and bidirectional long short-term memory network using FEMTO-ST bearing data set. The rotation speed of the system, the radial force on the bearing and the load torque are the main parameters of the operation. In order to ensure the comparability of the experiments, these three parameters are kept constant during each measurement time in the experiments. At the same time, in all experiments, the temperature is kept at about 45°.
[0036] The data set of a certain working condition is used as the vibration data of the actual processing, and other data sets are selected as the vibration data of the bearing processing measured in the laboratory with the same or similar working conditions as the actual processing. The selected laboratory data is used as the input data set.
[0037] Improved exponential smoothing is used to reduce the noise of the signal, and then a time window is introduced to enhance the features. The exponential smoothing process adopts a weighted observation value calculation method proposed in the Python data analysis library Pandas, that is, each coefficient is a general weighted average of the exponential function:
[0038] 4) Using the equivalence between dropout technology and Bayesian inference, in L 2 Under regular conditions, the objective function is optimized as:
[0039] 5) Build a Bi-LSTM network (bidirectional long short-term memory network) structure and put the processed data into the network to obtain the output result.
Claims
1. A mechanical bearing life prediction method based on exponential smoothing and bidirectional long short-term memory network, characterized in that: include: Obtain historical operation and maintenance data of the equipment, and organize and archive the equipment data in the form of pre-processed data; Performing data cleaning and feature preprocessing on the device data to obtain processed data; The processed data is smoothed and then input into the trained Bi-LSTM to obtain an output result, which is the mechanical bearing life prediction result; Training methods for exponential smoothing and Bi-LSTM networks: The acquired equipment historical operation and maintenance data are subjected to exponential smoothing, and then a time window is introduced to enhance the features; Build a Bi-LSTM network structure, which includes several Bi-LSTMs and N fully connected layers connected in sequence; Divide the sample data into training set and test set; Set up the training plan: how many times to loop over the training set, the number of training batches, random shuffling, optimizer, loss function, and saving conditions; Inputting the training set into the Bi-LSTM network and training according to the training scheme to obtain a preliminary training model; The exponential smoothing process and each Bi-LSTM block include sequentially connected units, BN layers and ReLU activation functions; The exponential smoothing process uses a weighted observation value calculation method proposed in the Python data analysis library Pandas, that is, each coefficient is a general weighted average of the exponential function: For any given α, in calculating y t When x is far away from the current time t-i The weight is small and its influence on the calculation can be ignored, so it can be understood as x within a certain range s at the near end. t-i Plays a major role, namely {x t-i |0≤i≤s-1}, s is regarded as the smoothing coefficient, and the relationship between α and s is as follows: The mathematical transformation formula of the Bi-LSTM block is include: h t =f(w1x t+ w2h t-1 ) h′ t =f(w3x t +w5h′ t+1 ) O t =g(w4h t +w6h′ t ) Where: h t , h′ t are the outputs of the forward and backward propagation layers and the backward propagation layer at time t; w1 and w3 are the weight matrices of the input layer and the backward propagation layer respectively; w3 and w5 are the weight matrices from the forward and backward propagation layers to their own propagation layers respectively; w4 and w6 are the weight matrices from the forward and backward propagation layers to the output layer respectively; t is the output value of the final output gate; function f(·) represents the cell calculation process; function g(·) represents the function for concatenating the forward and backward propagation results; The calculation of the BN layer includes: Among them, x p is the input of the pth batch, E[x p ] is x p The average value of Var[x p ] is x p The standard deviation of , γ and β are adaptively updated learnable parameters, y p is the output of the BN layer neurons.
2. The method for predicting mechanical bearing life based on exponential smoothing and bidirectional long short-term memory network according to claim 1 is characterized in that: The calculation of the ReLU activation function includes: Where x is the function input.