Hidden Markov prediction compensation method and system for gear wear evolution

Abstract

The invention provides a hidden Markov prediction compensation method and system for gear wear evolution, and belongs to the technical field of gear transmission, and the method comprises the steps: constructing a hidden Markov chain of a gear wear state-compensation parameter, building an encoder coding-decoding frame in a gear compensation system, extracting a hidden Markov chain state sequence by using an encoder, and carrying out the coding-decoding of the hidden Markov chain state sequence. A decoder is used for feeding back a prediction result to the compensation module; the observation sequence of the hidden Markov chain is substituted into the trained hidden Markov model, a corresponding state sequence is obtained, and life end point prediction is carried out; the state sequence is converted into an end-of-life prediction curve, and an optimal end-of-life value with the shortest distance between a prediction value and a target is searched in a pre-drawn compensation optimization solution space to serve as a reference wear state; and the reference wear state is substituted into an encoder in the hidden Markov chain, encoding curve parameters are obtained, parameter optimization is carried out on a compensation curve, and a traditional passive response strategy is converted into an active prevention strategy.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of gear transmission, in particular to a gear wear state prediction and compensation method and system based on a hidden Markov model. BACKGROUND

[0002] As a key component in mechanical transmission systems, the wear state of gears directly affects the operating efficiency and service life of equipment. With the development of Industry 4.0 and intelligent manufacturing, higher requirements are placed on the reliability and life of gear transmission systems. Traditional gear wear monitoring methods mainly rely on periodic inspection and experience-based judgment, making it difficult to achieve accurate prediction and timely compensation.

[0003] In the prior art, gear wear monitoring mainly uses methods such as vibration analysis, oil analysis, and acoustic emission to obtain wear information. For example, by analyzing the spectral characteristics of vibration signals, the wear degree of gears is identified; or by detecting the concentration of metal particles in lubricating oil, the wear state is evaluated. However, these methods can only reflect the current wear state and lack the ability to predict future wear evolution trends.

[0004] In terms of wear compensation, the prior art mainly adopts a passive response strategy, i.e., adjustments are made only after detecting that the wear exceeds a threshold. This method often lags behind actual needs and is difficult to achieve optimal compensation effect. In recent years, although some research has attempted to apply machine learning methods to gear wear prediction, there are problems such as insufficient model accuracy, poor adaptability, and inability to effectively combine with compensation strategies.

[0005] In particular, traditional gear wear prediction methods are difficult to handle the randomness and uncertainty in the wear process, and compensation strategies are often fixed empirical formulas that are difficult to adaptively adjust according to the dynamic changes in wear state. This makes the existing technology unable to achieve accurate prediction and optimal compensation of gear wear, limiting the service life and reliability of equipment. SUMMARY

[0006] In view of the above deficiencies of the prior art, the present application aims to provide a hidden Markov prediction and compensation method and system for gear wear evolution, which can accurately predict the evolution trend of the wear state of gears and implement an optimal compensation strategy based on the prediction results, thereby prolonging the service life of gears and improving system reliability.

[0007] The present application proposes a hidden Markov prediction and compensation method for gear wear evolution, comprising:

[0008] A hidden Markov chain of gear wear state-compensation parameters is constructed, including: establishing an encoder-encoder framework in the gear compensation system, using an encoder to extract a hidden Markov chain state sequence, and using a decoder to feed back the prediction results to the compensation module;

[0009] The observation sequence of the hidden Markov chain is substituted into the trained hidden Markov model to obtain its corresponding state sequence and perform life end point prediction;

[0010] The state sequence is converted into a life end point prediction curve, and the optimal life end point value closest to the target distance in the pre-drawn compensation optimization solution space on the life end point prediction curve is found as the reference wear state;

[0011] The reference wear state is substituted into the encoder in the hidden Markov chain to obtain the encoding curve parameters and perform parameter optimization on the compensation curve.

[0012] As a preferred, in the hidden Markov chain of constructing gear wear state-compensation parameters, the encoder is composed of n LSTM neural networks and n+1 fully connected layers, the hidden layer output is the encoding curve and its probability distribution, where i is the i th data of the input data sequence, i is also the state of the hidden Markov chain corresponding to the i th input data; the fully connected layer includes an input layer, a hidden layer and an output layer, and the output is whether the state is updated, if updated, it is 1, otherwise it is 0; the decoder is composed of a neural network and a cross-entropy loss function, the decoder and the encoder are related in weight and bias, the prediction result is fed back to the compensation module, the cross-entropy loss of the prediction value and the target value is calculated, and the hidden Markov model is trained and updated.

[0013] As a preferred, the encoder maps the gear wear state and compensation parameters in real time, obtains the n key performance indicators closest to the compensation state relationship under the gear wear state through training, and gives different weight values according to their importance.

[0014] As a preferred, the calculation of the encoder includes:

[0015] The encoder divides the input data into multiple states, obtains the curve and its probability distribution under each state through n neural networks, and the neural networks take the key performance indicators of each state as input and the probability distribution of the corresponding state as output;

[0016] State update: input the state encoding curve and its probability distribution into the decoder, and take whether the state is updated as output, whether the state is updated is determined by its probability distribution;

[0017] State sequence prediction: substitute the prediction result and the true result into the hidden Markov model, calculate the cross-entropy loss of the prediction value and the true value, and update the model.

[0018] As a preferred, the calculation of the decoder includes:

[0019] The decoder inputs the state output by the hidden Markov model and the state output by the state updating module into the neural network, normalizes the conditional distribution matrix of each state, and accumulates the results to obtain a hidden Markov chain transition probability matrix and an observation probability matrix, and calculates a state sequence with the maximum probability.

[0020] As preferred, in the step of substituting the observation sequence of the hidden Markov chain into the trained hidden Markov model, the state sequence is substituted into the trained hidden Markov model to generate an output sequence, which is compared with the observation sequence. If they are completely identical, the model is a correct model. Otherwise, the cross-entropy loss of the probability matrix is continuously calculated, the model is updated, and the process is repeated until the cross-entropy loss is less than a threshold value.

[0021] As preferred, the future possible state of the hidden Markov chain is predicted according to the trained hidden Markov model, the observation curve under different states is used for life end point prediction, the life end points under different states are substituted into the hidden Markov chain model to obtain the probability distribution of each life end point, the probability distribution of the life end point is substituted into the cross-entropy loss function, the life end point corresponding to the maximum value of the probability distribution of the life end point is taken as the predicted life end point value, and the probability of the predicted life end point value is greater than 80%.

[0022] As preferred, the life end point prediction curve represents a historical curve and a prediction curve of the wear state evolution in the life cycle, the life cycle is divided according to time, and the prediction curve of the life end point represents the probability distribution curve of each life end point prediction in the life cycle.

[0023] As preferred, in the step of substituting the reference wear state into the encoder in the hidden Markov chain:

[0024] Through the cross-entropy loss function of the life end point prediction, the life end point corresponding to the maximum value of the probability distribution of the predicted life end point is found, which is the optimal solution. If the probability of the optimal solution is less than 80%, the optimization is continued. Otherwise, the optimal solution is output.

[0025] The output corresponding to the optimal solution is taken as the reference wear state, the reference wear state curve is input into the encoder to obtain each key performance indicator under the wear state, and the parameters corresponding thereto, i.e., the compensation parameters under the predicted wear state, are found.

[0026] The gear wear evolution hidden Markov prediction compensation system comprises:

[0027] State-parameter mapping module: Establishes a hidden Markov chain of gear wear state-compensation parameters, integrates an acoustic emission sensor to collect friction signals on the gear surface inside the encoder, divides the input data into multiple states by the encoder, obtains the curve and its probability distribution in each state through n neural networks, and feeds back the prediction result to the compensation module through the mutual correlation of the weights and biases of the encoder and the decoder to calculate the cross-entropy loss of the predicted value and the target value, and train the encoder and the decoder.

[0028] End-of-life prediction module: The output of the hidden Markov model is substituted into the trained model to generate an output sequence, which is compared with the observed sequence until they are identical.

[0029] End-of-life prediction curve: The historical curve and the predicted curve of the wear state evolution in the life cycle are obtained by the encoder output, the end-of-life in different states is substituted into the hidden Markov model to obtain the probability distribution of each end-of-life, and the probability distribution of the end-of-life is substituted into the cross-entropy loss function, and the end-of-life corresponding to the maximum value of the probability distribution of the end-of-life is taken as the predicted end-of-life value to output the optimal solution.

[0030] Parameter optimization module: The predicted end-of-life value is evolved in the life cycle, and the optimal end-of-life value closest to the target distance is found in the compensation optimization solution space pre-drawn on the life prediction curve as the reference wear state to output the optimal solution; the optimal solution is substituted into the hidden Markov model to find the observed value and state corresponding to the optimal solution, and the observed value and state are put into the model to update the model, and if the updated cross-entropy loss is less than the set threshold, the optimal solution is output, otherwise, the previous step is repeated until the updated model meets the threshold condition.

[0031] The beneficial effects of the present application include:

[0032] 1. By constructing a hidden Markov chain of gear wear state-compensation parameters, accurate mapping from observation signals to wear states is realized, and the prediction accuracy is improved by about 40% compared with traditional methods.

[0033] 2. The combination of multi-layer LSTM neural network and hidden Markov model effectively captures the time sequence characteristics and state transition rules of gear wear, and reduces the end-of-life prediction error to less than 1 / 3 of traditional methods.

[0034] 3. Based on the predicted wear state, an active compensation strategy is implemented to change passive response to active prevention, and the service life of the gear is extended by 30%-50%.

[0035] 4. By adaptively optimizing the compensation parameters, the system can cope with different working conditions and load conditions, maintain the best operating state, and reduce unplanned shutdown and emergency failures by more than 60%.

[0036] 5. Modular design makes the system has good scalability and adaptability, suitable for various gear transmission systems, investment recovery period is not more than 18 months. BRIEF DESCRIPTION OF DRAWINGS

[0037] Figure 1 The flow chart of the gear wear evolution hidden Markov prediction compensation method of the application;

[0038] Figure 2 The structural diagram of the encoder-decoder framework in the application;

[0039] Figure 3 The example diagram of the end-of-life prediction curve in the application;

[0040] Figure 4 The schematic diagram of the compensation optimization solution space in the application;

[0041] Figure 5 The structural block diagram of the gear wear evolution hidden Markov prediction compensation system of the application. DETAILED DESCRIPTION

[0042] Please refer to Figures 1-5 , the specific embodiments of the application will be described in detail below with reference to the drawings. It should be noted that those skilled in the art should understand that these embodiments are only used to illustrate the application, and should not be regarded as limiting the scope of the application.

[0043] The application provides a gear wear evolution hidden Markov prediction compensation method and system, which realizes accurate prediction of gear wear state and optimization adjustment of compensation parameters by constructing a hidden Markov chain of gear wear state-compensation parameters.

[0044] Referring to Figure 1 , the gear wear evolution hidden Markov prediction compensation method of the application includes four main steps: constructing a hidden Markov chain of gear wear state-compensation parameters; substituting the observation sequence of the hidden Markov chain into the trained hidden Markov model to obtain its corresponding state sequence and perform end-of-life prediction; converting the state sequence into an end-of-life prediction curve, and finding the optimal end-of-life value in the pre-drawn compensation optimization solution space; substituting the reference wear state into the encoder in the hidden Markov chain to obtain the encoding curve parameters and optimize the parameters of the compensation curve.

[0045] Referring to Figure 5 , the gear wear evolution hidden Markov prediction compensation system of the application includes four main modules: state-parameter mapping module 1, end-of-life prediction module 2, end-of-life prediction curve module 3, and parameter optimization module 4.

[0046] The specific embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0047] In a preferred embodiment of the present application, as shown in Figure 2 The hidden Markov chain of gear wear state-compensation parameters is implemented by establishing an encoder- decoder framework in the gear compensation system. Specifically, the encoder is used to extract the hidden Markov chain state sequence, and the decoder is used to feed back the prediction results to the compensation module.

[0048] The encoder is composed of n LSTM (Long Short-Term Memory) neural networks and n+1 fully connected layers. Among them, the value of n is usually determined according to the complexity of the gear wear state. In practical applications, n is usually taken as 3 to 5 to balance the model complexity and prediction accuracy. For example, for a wind turbine gearbox, n = 4 can achieve good results; for an automobile gearbox, n = 3 can meet the demand.

[0049] In the encoder, each LSTM neural network is responsible for extracting features at different time scales. The first level LSTM network extracts short time scale features (such as microscopic friction features within 0.1 seconds), and the last level LSTM network extracts long time scale features (such as macroscopic wear trends within 10 minutes). The output of the LSTM network is the hidden layer state, denoted as encoding curves H1 to and their probability distributions.

[0050] Mathematically, for the i-th data point of the input data sequence, the output of the LSTM network can be represented as:

[0051] ,

[0052] Where: is the hidden state at the current time, is the current input, is the hidden state at the previous time, is the cell state at the previous time.

[0053] The specific structure of the LSTM network includes an input gate, a forget gate, and an output gate, and its calculation process is as follows:

[0054] ,

[0055] ,

[0056] ,

[0057] ,

[0058] ,

[0059] ,

[0060] where: is the output vector of the forget gate, ranging from [0, 1], controlling the retention degree of the previous state; is the output vector of the input gate, ranging from [0, 1], controlling the acceptance degree of new input; is the candidate cell state vector, ranging from [-1, 1]; is the current cell state vector; is the output vector of the output gate, ranging from [0, 1], controlling the output degree of the cell state to the outside; is the current hidden state vector; is the sigmoid activation function, mapping the input to the interval [0, 1]; , , , are the weight matrices of the forget gate, input gate, cell state, and output gate, respectively; , , , are the corresponding bias vectors, respectively; represents the element-wise multiplication operation.

[0061] In the gear system, these gating mechanisms enable LSTM to selectively remember long-term wear trends (such as maintaining long-term states through the forget gate) and respond to short-term incidents (such as receiving new data through the input gate). For example, when the gear experiences a sudden wear acceleration, the input gate may increase its value, making the network pay more attention to the current input.

[0062] The fully connected layer in the encoder includes an input layer, a hidden layer, and an output layer. The input layer receives the output of the LSTM network, the hidden layer performs feature transformation, and the output layer generates a state update signal. When the state needs to be updated, the output is 1, otherwise it is 0. The calculation process of the fully connected layer is as follows:

[0063] ,

[0064] where: is the output vector, representing the state update signal; is the input vector, which is the output of the LSTM network; is the weight matrix, whose dimensions depend on the dimensions of the input and output; is the bias vector; is the activation function (usually using ReLU or sigmoid function). In gear wear monitoring, the ReLU function is defined as , which helps to extract nonlinear features; the sigmoid function is defined as , adapted to generate a state update signal between 0 and 1.

[0065] The decoder is composed of a neural network and a cross-entropy loss function. The decoder is associated with the encoder in terms of weights and biases, ensuring that the two can work together. The decoder feeds back the prediction results to the compensation module and calculates the cross-entropy loss between the predicted value and the target value for the training update of the hidden Markov model.

[0066] The cross-entropy loss function is defined as:

[0067] ,

[0068] Where: is the loss value, which is a scalar; is the true label, representing the true state category in gear wear state classification (e.g. normal wear is [1, 0, 0], accelerated wear is [0, 1, 0]); is the predicted probability, representing the probability of each state predicted by the model; is the number of samples; is the natural logarithm function. The smaller the cross-entropy loss, the closer the prediction result is to the true situation.

[0069] In practical applications, the encoder maps the gear wear state and compensation parameters in real time, obtains the n key performance indicators that are most closely related to the compensation state under the gear wear state through training, and assigns different weight values according to their importance. These key performance indicators may include friction coefficient of tooth surface, vibration spectrum characteristics, temperature change rate, etc.

[0070] For example, for a wind turbine gearbox, the key performance indicators may include: meshing frequency vibration amplitude (weight 0.35), friction coefficient of tooth surface (weight 0.25), noise spectrum energy (weight 0.2), temperature gradient (weight 0.1), and lubricating oil particle concentration (weight 0.1). These weight values are determined through statistical analysis of a large amount of historical data and expert experience, and can reflect the importance of each indicator in wear state judgment.

[0071] The calculation process of the encoder specifically includes three steps: first, the encoder divides the input data into multiple states, and obtains the curve and its probability distribution under each state through n neural networks. The neural network takes the key performance indicators of each state as input and the probability distribution of the corresponding state as output. Second, update the state, input the state encoding curve and its probability distribution into the decoder, and take whether to update each state as output. Whether to update the state is determined by its probability distribution. Finally, predict the state sequence, substitute the prediction result and the true result into the hidden Markov model, calculate the cross-entropy loss between the predicted value and the true value, and update the model.

[0072] The decoder's computation process includes: inputting the states output by the Hidden Markov Model and the states output by the state update module into the neural network; normalizing the conditional distribution matrices of each state and accumulating the results to obtain the Hidden Markov Chain transition probability matrix A and the observation probability matrix B; and finally calculating the state sequence with the highest probability.

[0073] The transition probability matrix A represents the probability of transitioning from one state to another, and its elements are... This represents the probability of transitioning from state i to state j:

[0074] ,

[0075] in: The state transition probability matrix has dimension 1. , The number of states; The element in the i-th row and j-th column of the matrix represents the probability of transitioning from state i to state j. Let represent the conditional probability that the state is j at time t+1, given that the state is i at time t. This represents the state variable at time t. For a gear system, the state may include normal wear, light wear, moderate wear, and heavy wear.

[0076] The observation probability matrix B represents the probability of observing a specific observation under a specific state, and its elements are... This represents the probability of observing observation k in state j:

[0077] ,

[0078] in: The observation probability matrix has dimensions of . , For the number of states, The number of observations; The first in the matrix The element in the k-th row represents the probability of observing observation k under state condition; Let represent the conditional probability of observing the observation value k given that the state is i at time t. This represents the observed variable at time t. In a gear system, the observed value might be a vibration signal measured by a sensor, a temperature change, etc.

[0079] In practical applications, the initial values ​​of the transition probability matrix A and the observation probability matrix B are usually set based on historical data or prior knowledge, and then continuously optimized through model training. For example, for a newly installed gear system, the initial state may mainly focus on the normal wear area, so the probability of transitioning from a normal state to a normal state in the transition probability matrix is ​​relatively high (e.g., 0.95), while the probability of transitioning from a normal state to an abnormal state is relatively low (e.g., 0.05). As the system runs for longer, these probabilities are continuously adjusted based on actual observation data.

[0080] After obtaining the state sequence, this invention substitutes the state sequence into the trained Hidden Markov Model to generate an output sequence, which is then compared with the observed sequence. If the two are completely identical, the model is considered correct; otherwise, the cross-entropy loss of the probability matrix is ​​calculated, and the model is updated until the cross-entropy loss is less than a threshold.

[0081] In a preferred embodiment, the threshold is set to 0.01, an empirical value derived from extensive experiments. When the cross-entropy loss is less than 0.01, the model's prediction accuracy typically reaches over 95%, meeting the needs of engineering applications. A threshold that is too small will lead to excessively long training times, while a threshold that is too large may affect prediction accuracy.

[0082] Based on a trained Hidden Markov Model (HMM), this invention predicts the possible future states of a HMM chain and performs lifetime prediction on the observed curves under different states. Specifically, the lifetime endpoints under different states are substituted into the HMM model to obtain the probability distribution of each lifetime endpoint; then, the probability distribution of the lifetime endpoints is substituted into the cross-entropy loss function, and the lifetime endpoint corresponding to the maximum value of the probability distribution is taken as the predicted lifetime endpoint value. This invention requires the probability of the predicted lifetime endpoint value to be greater than 80% to ensure the reliability of the prediction results.

[0083] In mathematics, life end prediction can be expressed as:

[0084] ,

[0085] in: The predicted end-of-life value is a point in time; The candidate lifetime endpoint indicates the possible end-of-life time. For a given observation sequence Next life end The posterior probability; This represents the lifespan end point with the highest probability. This represents the observation sequence from time 1 to time t. In a gear system, these observations might be data sequences of vibration, temperature, etc., measured by sensors.

[0086] In practical applications, the life end prediction curve represents the historical curve and the prediction curve of the wear state evolution in the life cycle. The life cycle is divided according to time, usually in units of working periods or maintenance periods of the equipment. The prediction curve of the life end represents the probability distribution curve of each life end prediction in the life cycle, as shown in Figure 3 .

[0087] For example, for a wind turbine gearbox, the life cycle can be divided into an initial running-in period (0-1000 hours), a normal operation period (1000-20000 hours), and a degradation period (more than 20000 hours). In the normal operation period, the system predicts the possible life end of the gear, as shown in Figure 3 , where the abscissa represents time and the ordinate represents the probability of becoming the life end at the corresponding time point. As can be seen from the figure, there is a probability peak near t=15000 hours, indicating that the probability of becoming the life end at this time is the largest, with a probability of about 0.85, which exceeds the 80% threshold set by the present application, so the system takes t=15000 hours as the predicted life end value.

[0088] After obtaining the state sequence, the present application converts it into a life end prediction curve, as shown in Figure 3 . In the life end prediction curve, a compensation optimization solution space is pre-drawn, as shown in Figure 4 . The compensation optimization solution space is a multi-dimensional space, where each dimension corresponds to a compensation parameter, and each point represents a specific set of compensation parameters.

[0089] In the compensation optimization solution space, the present application finds the optimal life end value closest to the prediction value and the target distance as the reference wear state. Specifically, by using the cross-entropy loss function of the life end prediction, the life end corresponding to the maximum value of the probability distribution of the predicted life end is found, which is the optimal solution. If the probability of the optimal solution is less than 80%, the optimization continues; otherwise, the optimal solution is output.

[0090] In an actual application case, for a certain type of wind turbine gearbox, the system predicts multiple possible life ends, with a probability of 0.85 at t=15000 hours, a probability of 0.10 at t=18000 hours, and a probability of 0.05 at t=12000 hours. Since the probability of t=15000 hours exceeds the 80% threshold, the system takes it as the optimal life end and formulates a compensation strategy based on it.

[0091] The selection criterion for the optimal end-of-life is set at 80%, based on engineering experience and risk management considerations. A higher probability threshold (e.g., 95%) would result in a system that is overly conservative, potentially missing some effective compensation opportunities; while a lower threshold (e.g., 60%) could introduce excessive uncertainty, reducing the effectiveness of compensation. The 80% threshold strikes a good balance between reliability and flexibility.

[0092] Finally, the system uses the output corresponding to the optimal solution as the reference wear state and inputs the reference wear state curve into the encoder to obtain the key performance indicators at that wear state and find the corresponding parameters, i.e., the compensation parameters at the predicted wear state.

[0093] Specifically, the encoder maps the reference wear state to the key performance indicator space through forward propagation, obtaining a set of key performance indicator values. Then, the system looks up the most matching compensation parameter combination in the pre-established performance indicator-compensation parameter mapping table. This mapping relationship is trained through a large amount of historical data and can reflect the optimal compensation strategy under different wear states.

[0094] For example, for a certain wind turbine gearbox, the key performance indicators corresponding to the reference wear state determined by the system may be: meshing frequency vibration amplitude 0.35 mm / s, tooth surface friction coefficient 0.12, noise spectrum energy 25 dB, temperature gradient 5℃ / m, and lubricating oil particle concentration 15 ppm. Based on these indicators, the system finds the optimal compensation parameters: increase the gear preload by 5%, increase the lubricating oil flow by 8%, and reduce the operating speed by 3%. This set of compensation parameters can effectively delay the wear process and prolong the service life of the gear.

[0095] The optimization of compensation parameters is an iterative process. After the system implements compensation, it continues to monitor the actual operating state of the gear and compares it with the predicted state. If there is a significant deviation between the two, the system will re-evaluate the wear state, update the hidden Markov model, and adjust the compensation strategy. This closed-loop feedback mechanism ensures that the system can adapt to changing working conditions and maintain the best compensation effect.

[0096] Reference Figure 5 The gear wear evolution hidden Markov prediction compensation system of the present application includes four main modules: state-parameter mapping module 1, end-of-life prediction module 2, end-of-life prediction curve module 3, and parameter optimization module 4.

[0097] The state-parameter mapping module 1 is responsible for establishing the hidden Markov chain of gear wear state-compensation parameters. In this module, an acoustic emission sensor is integrated inside the encoder for collecting friction signals on the gear surface. The encoder divides the input data into multiple states, and through n neural networks, it obtains the curve and its probability distribution in each state. The decoder is correlated with the encoder in terms of weights and biases, and the prediction results are fed back to the compensation module to calculate the cross-entropy loss between the predicted value and the target value, and to train the encoder and decoder.

[0098] In a preferred embodiment, the acoustic emission sensor uses a piezoelectric sensor with a working frequency range of 50 kHz-500 kHz and a sensitivity of 75 dB±3 dB (reference 0 dB=1V / μbar). This sensor can effectively capture high-frequency acoustic signals generated during the friction process on the gear surface, providing a reliable data source for wear state recognition.

[0099] The structure of the neural network preferably uses an LSTM network, with each LSTM unit containing 128 neurons and using tanh as the activation function. This structure can effectively capture the long-term dependencies of time series data and is suitable for evolution prediction of gear wear states.

[0100] During training, the learning rate is initially set to 0.001, the Adam optimizer is used, the batch size is 32, and the number of training rounds is 100. These parameter settings are based on empirical values obtained from extensive experiments, which can achieve a good balance between training efficiency and model performance.

[0101] The end-of-life prediction module 2 is responsible for inputting the output of the hidden Markov model into the trained model to generate an output sequence, which is compared with the observed sequence until they are identical.

[0102] In practical applications, the end-of-life prediction module uses an improved Viterbi algorithm to calculate the most likely state sequence. The core idea of the Viterbi algorithm is dynamic programming, which recursively calculates the maximum probability path to find the most likely state sequence for a given observation sequence.

[0103] The improved Viterbi algorithm takes into account the historical dependence of state transitions, not only depending on the previous state but also considering longer historical state sequences. This improvement can more accurately capture the complex evolution of gear wear and improve prediction accuracy.

[0104] The time complexity of the algorithm is O(N²T), where N is the number of states and T is the sequence length. In practical applications, N usually does not exceed 10, and T can reach thousands or more, so the efficiency of the algorithm is mainly affected by the sequence length. Through optimized implementation, the algorithm can run in real time on ordinary industrial computing platforms, meeting the needs of practical applications.

[0105] The life end prediction curve module 3 is responsible for obtaining the historical curve and the prediction curve of the wear state evolution in the life cycle by encoding the output of the encoder, substituting the life end in different states into the hidden Markov model, obtaining the probability distribution of each life end, and substituting the probability distribution of the life end into the cross-entropy loss function. The life end corresponding to the maximum of the probability distribution of the life end is taken as the predicted life end value, and the optimal solution is output.

[0106] In practical applications, the life end prediction curve is usually represented by a smooth curve to reduce the influence of noise and fluctuations. The smoothing method can use moving average or exponential smoothing, and the smoothing parameter is usually set to 0.1-0.3, which can filter out short-term fluctuations while retaining trend information.

[0107] The probability distribution of the life end usually presents a unimodal or multimodal pattern. Unimodal distribution indicates that the system's prediction of the life end is relatively concentrated, while multimodal distribution indicates that there are multiple possible life ends. In the case of multimodal distribution, the system selects the peak value with the highest probability as the prediction result, provided that the probability of the peak value exceeds the 80% threshold.

[0108] The parameter optimization module 4 is responsible for finding the optimal life end value closest to the target distance in the compensation optimization solution space pre-drawn on the life prediction curve based on the wear state evolution of the predicted life end value in the life cycle, outputting the optimal solution as the reference wear state; the optimal solution is substituted into the hidden Markov model to find the observation value and state corresponding to the optimal solution, which is put into the model to update the model. If the updated cross-entropy loss is less than the set threshold, the optimal solution is output, otherwise repeat the previous step until the updated model meets the condition of less than the threshold.

[0109] In practical applications, the compensation optimization solution space is usually a multi-dimensional space, each dimension corresponding to a compensation parameter. For example, for a wind turbine gearbox, the compensation parameters may include pre-tightening force, lubricating oil flow, running speed, etc. Each point in the space represents a specific set of compensation parameters, and the color or size of the point represents the expected effect of the setting.

[0110] The optimization process uses gradient descent or simulated annealing algorithm to search for the optimal solution in the solution space. The learning rate of the gradient descent algorithm is usually set to 0.01-0.1, and the iteration number is 100-500 times. The initial temperature of the simulated annealing algorithm is usually set to 100, the cooling rate is 0.95, and the iteration number is 1000 times. These parameter settings can achieve a good balance between optimization efficiency and result quality.

[0111] The updated cross-entropy loss threshold is usually set to 0.01, consistent with the threshold in the aforementioned life end prediction module. This ensures that the model updating and prediction processes use a unified evaluation standard, maintaining the consistency of the system.

[0112] To illustrate the implementation effect of the present application more intuitively, a specific application example is given below.

[0113] A certain wind farm gearbox has wear problems after long-term operation. Traditional maintenance methods cannot accurately predict wear trends, leading to unexpected downtime and high maintenance costs. After applying the method of the present application, the system can monitor the gear wear state in real time, predict future wear trends, and implement proactive compensation strategies.

[0114] The specific implementation process is as follows:

[0115] 1. Install acoustic emission sensors at key positions of the gearbox to collect gear face friction signals. The signal sampling rate is set to 1MHz, and a data packet is formed every 10 minutes for processing.

[0116] 2. Build an encoder containing 4 LSTM networks and a corresponding decoder. The number of layers of the LSTM network is 2, and each layer has 128 neurons. The fully connected layer uses the ReLU activation function, and the output layer uses the sigmoid activation function.

[0117] 3. Use historical data (including normal operation data and known wear cases) to train the hidden Markov model. The initial state includes normal wear, mild wear, moderate wear, and severe wear. The training process uses the Adam optimizer with a learning rate of 0.001, a batch size of 32, and 100 training rounds.

[0118] 4. The system monitors the running state of the gearbox in real time, and substitutes the observed data into the trained model to predict the future wear trend. For example, in a certain prediction, the system gives the possible lifetime endpoint distribution: the probability of t=3000 hours is 0.05, the probability of t=5000 hours is 0.85, and the probability of t=7000 hours is 0.10. Since the probability of t=5000 hours exceeds the threshold of 80%, the system takes it as the predicted lifetime endpoint.

[0119] 5. Based on the prediction result, the system searches for the optimal compensation strategy in the compensation optimization solution space. After optimization, the system determines the optimal compensation parameters: pre-tightening force increases by 5%, lubricating oil flow increases by 8%, and operating speed decreases by 3%.

[0120] 6. After implementing the compensation strategy, the system continues to monitor the deviation between the actual wear state and the predicted state. If the deviation exceeds the threshold, the wear state is re-evaluated, the model is updated, and the compensation strategy is adjusted.

[0121] The application result shows that compared with the traditional method, the service life of the wind power gear box is prolonged by about 45%, the number of unplanned shutdown is reduced by 70%, and the maintenance cost is reduced by 50%. The investment recovery period is about 14 months, and the economic benefit is remarkable.

[0122] The application provides a gear wear evolution hidden Markov prediction compensation method and system, realizes accurate prediction of gear wear state and optimal adjustment of compensation parameters by constructing a hidden Markov chain of gear wear state-compensation parameters.

[0123] The core innovation of the method is that a prediction framework combining a multi-layer LSTM network and a hidden Markov model is constructed, which can effectively capture the time sequence characteristics and state transition rules of gear wear; a life end prediction method based on probability distribution is designed, which can quantify the uncertainty of prediction and improve the reliability of decision-making; and an active compensation strategy based on prediction is established, which can respond to wear deterioration in advance and prevent problems from occurring.

[0124] The application is suitable for various gear transmission systems, including but not limited to wind power equipment, industrial machinery, automobile gearboxes and the like.

[0125] The above-described embodiments only express the specific implementation of the application, and the description is more specific and detailed, but it cannot be understood as a limitation on the scope of the patent of the application. It should be pointed out that for ordinary skilled persons in the art, without departing from the concept of the application, a number of modifications and improvements can be made, which belong to the protection scope of the application.

Claims

1. A hidden Markov prediction compensation method for gear wear evolution, characterized in that, The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain.

2. The hidden Markov prediction compensation method for gear wear evolution according to claim 1, characterized in that, The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain.

3. The hidden Markov prediction compensation method for gear wear evolution according to claim 2, characterized in that, The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain.

4. The hidden Markov prediction compensation method for gear wear evolution according to claim 3, characterized in that, The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain.

5. The hidden Markov prediction compensation method for gear wear evolution according to claim 4, characterized in that, The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application relates to a gear wear state-compensation parameter hidden Markov chain, and a method for constructing the gear wear state-compensation parameter hidden Markov chain. The application 6. The hidden Markov prediction compensation method for gear wear evolution according to claim 5, characterized in that, The step of substituting the observation sequence of the hidden Markov chain into the trained hidden Markov model substitutes the state sequence into the trained hidden Markov model to generate an output sequence and compare the output sequence with the observation sequence, and if the two sequences are completely identical, the model is a correct model; otherwise, the cross-entropy loss of the probability matrix is continuously calculated, the model is updated, and the process is repeated until the cross-entropy loss is less than a threshold.

7. The hidden Markov prediction compensation method for gear wear evolution according to claim 6, characterized in that, According to the trained hidden Markov model, the future possible states of the hidden Markov chain are predicted, the observation curves under different states are used for life end point prediction, the life end points under different states are substituted into the hidden Markov chain model to obtain the probability distribution of each life end point, the probability distribution of the life end point is substituted into the cross-entropy loss function, and the life end point corresponding to the maximum value of the probability distribution of the life end point is taken as a predicted life end point value, and the probability of the predicted life end point value is greater than 80%.

8. The hidden Markov prediction compensation method for gear wear evolution according to claim 7, characterized in that, The life end point prediction curve represents a history curve and a prediction curve of wear state evolution in a life cycle, the life cycle is divided according to time, and the prediction curve of the life end point represents a probability distribution curve of each life end point prediction in the life cycle.

9. The hidden Markov prediction compensation method for gear wear evolution according to claim 8, characterized in that, In the step of substituting the reference wear state into the encoder in the hidden Markov chain: An optimal solution is found by using the cross-entropy loss function of the life end point prediction, and the optimal solution is the life end point corresponding to the maximum value of the probability distribution of the predicted life end point; if the probability of the optimal solution is less than 80%, the optimization is continued; otherwise, the optimal solution is output; The output corresponding to the optimal solution is taken as the reference wear state, the reference wear state curve is input into the encoder to obtain each key performance indicator under the wear state, and the parameters corresponding to the key performance indicators, i.e., the compensation parameters under the predicted wear state, are found.

10. A Hidden Markov prediction compensation system for gear wear evolution, for implementing the Hidden Markov prediction compensation method for gear wear evolution according to any one of claims 1 to 9, characterized in that, The system comprises: A state-parameter mapping module: a hidden Markov chain of gear wear state-compensation parameters is established, an acoustic emission sensor is integrated in the encoder to collect friction signals of a gear surface, the encoder divides input data into multiple states, n neural networks are used to obtain curves and probability distributions under each state, a decoder is associated with the encoder in terms of weight and bias, a prediction result is fed back to a compensation module, a cross-entropy loss of a prediction value and a target value is calculated, and the encoder and the decoder are trained; A life end point prediction module: the output of the hidden Markov model is substituted into a trained model to generate an output sequence, and the output sequence is compared with an observation sequence until the two sequences are completely identical; A life end point prediction curve: a history curve and a prediction curve of wear state evolution in a life cycle are obtained by using the output of the encoder, life end points under different states are substituted into a hidden Markov model to obtain probability distributions of each life end point, the probability distributions of the life end points are substituted into a cross-entropy loss function, a life end point corresponding to a maximum value of the probability distribution of the life end point is taken as a predicted life end point value, and an optimal solution is output. The parameter optimization module: in the life cycle, the wear state evolution is carried out to the predicted life end value, the optimal life end value closest to the target distance is found in the compensation optimization solution space pre-drawn on the life prediction curve as the reference wear state, and the optimal solution is output; the optimal solution is substituted into the hidden Markov model, the observation value and the state corresponding to the optimal solution are found, and the observation value and the state are put into the model to update the model, if the cross-entropy loss after updating is less than the set threshold, the optimal solution is output, otherwise, the above step is repeated until the updated model meets the condition of being less than the threshold.

Images