Data-driven nested model-based machine tool service performance degradation prediction method
By applying nested BP neural network model to predict performance degradation in machine tools, the problem of difficulty in monitoring and predicting machine tool performance degradation is solved, and accurate prediction and effective management of machine tool performance degradation process is achieved.
Patent Information
- Application Number
- CN202411863607.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-17
- Publication Date
- 2025-03-25
AI Technical Summary
The performance of the machine tool gradually deteriorates during service, resulting in an increase in the probability of unplanned failure downtime, making it difficult to effectively monitor and predict performance degradation through failure physical mechanism analysis.
The performance degradation prediction method based on nested BP neural network is adopted, and the performance characteristic parameters are extracted by monitoring the performance status of the machine tool, and the nested BP neural network model is used to fit and predict, track the performance degradation process, and determine the degree and trend of the degradation.
Accurate prediction of the machine tool performance degradation process is achieved, potential performance degradation modes can be discovered in advance, helping enterprises formulate suppression measures, extend the service life of the machine tool, and reduce maintenance costs.
Smart Images
Figure CN119692196A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the field of service performance research of numerical control machine tools, and is a performance degradation prediction method based on a nested model and targeting degradation data in an actual machine tool service process. Background Art
[0002] Machine tools are similar to human bodies. Their functions gradually decline over time, which is manifested in the gradual degradation of machine tool functions and performance, resulting in an increase in the rate of unplanned downtime and reduced service performance. By monitoring, analyzing and predicting the performance degradation of machine tools, enterprises can discover the potential performance degradation patterns of machine tools and develop reasonable suppression methods to avoid sudden failures of machine tools. This is of great significance for improving machine tool utilization, reducing machine tool maintenance costs, extending machine tool service life, and building a machine tool health status monitoring system.
[0003] Due to the complex structure and diverse failure modes of machine tools, it is difficult to study the performance degradation process of machine tools from the perspective of failure physical mechanism analysis. The external manifestation of performance degradation is that the output performance characteristic parameters of the machine tool gradually deviate from their normal range. This is because the specified functions of the machine tool are closely related to several output performance characteristic parameters. Using the external manifestations of performance degradation, a performance degradation modeling and analysis principle based on performance status monitoring is proposed. It monitors the performance status of the machine tool, obtains the performance signal, and extracts the performance characteristic parameters of the machine tool after signal processing and analysis; then the performance characteristic parameters are fitted and predicted to discover their time-varying laws, track the performance degradation process of the machine tool, and determine the degree and trend of the performance degradation of the machine tool.
[0004] Therefore, this patent proposes a performance degradation prediction method based on a nested neural network for degradation data during the actual service life of machine tools. This method has a certain degree of originality. Summary of the invention
[0005] The present invention proposes a data-driven performance degradation modeling prediction method, which obtains performance signals by monitoring the actual service performance status of the machine tool, and extracts the performance characteristic parameters of the machine tool after signal processing and analysis; adopts a nested BP neural network model and a hidden layer node number optimization method; and then fits and predicts the performance degradation characteristic parameters to discover their time-varying laws, track the performance degradation process of the machine tool, and determine the performance degradation degree and trend of the machine tool.
[0006] The present invention is realized by adopting the following technical means: a data-driven method for predicting the service performance degradation of machine tools based on nested models
[0007] S1 determines the performance degradation evaluation index, and establishes the performance degradation evaluation index based on the accuracy index and dynamic performance index of the machine tool performance degradation.
[0008] S2 obtains the performance signal of the machine tool through performance status monitoring. After signal processing, performance parameter extraction and fusion, it obtains information such as the intermediate status data and performance degradation data of the machine tool during service.
[0009] S3 Establishes a prediction model for machine tool service performance degradation
[0010] S3.1 Considering the BP model training error, training time, computational efficiency, and regression relationship, the network model is established. In view of the diversity of input data and the applicability of the algorithm, a cyclic calculation method is used in the design to select the optimal number of hidden layer nodes based on the mean square error of the training set.
[0011] S3.2 establishes a machine tool service performance degradation prediction model with a nested BP neural network, which is composed of two BP neural networks (BP neural network 1 and BP neural network 2). BP neural network 2 nests BP neural network 1.
[0012] S3.3 uses the training set to train the performance degradation prediction model, determines the weights of BP neural networks 1 and 2, and uses the test set to test the accuracy of the performance degradation model results. Perform accuracy analysis on the performance degradation prediction model and perform performance degradation prediction.
[0013] The advantages of the present invention are:
[0014] A performance degradation evaluation index based on the degradation of machine tool positioning accuracy and the first-order natural frequency degradation of the whole machine was established, which can accurately describe the service performance degradation process of the machine tool; a machine tool service performance degradation prediction model with a nested BP neural network was established, and the vibration signal and temperature signal of the machine tool were used as the intermediate transfer vectors of the model. A relationship model between the service time and working condition parameters and the machine tool accuracy degradation and dynamic performance degradation was constructed, which can obtain more accurate prediction results. BRIEF DESCRIPTION OF THE DRAWINGS
[0015] Figure 1 It is a flow chart of the implementation of the present invention;
[0016] Figure 2 It is a performance degradation prediction model based on nested BP neural network;
[0017] Figure 3 It is the network structure diagram, performance diagram and regression diagram of the nested BP neural network;
[0018] Figure 4 is the performance degradation prediction model accuracy analysis chart;
[0019] Figure 5 It is the service performance degradation prediction result diagram. DETAILED DESCRIPTION
[0020] The present invention is described in detail below with reference to the accompanying drawings and embodiments.
[0021] A data-driven nested model-based machine tool service performance degradation prediction method, the method comprising the following steps:
[0022] Step 1. Determination of performance degradation evaluation indicators
[0023] This patent establishes performance degradation evaluation indicators based on precision indicators and dynamic performance indicators of machine tool performance degradation.
[0024] The precision detection index of machine tools evaluates the performance degradation accuracy of machine tools by detecting the positioning accuracy of machine tools; positioning accuracy refers to the actual position accuracy of the main components of the machine tool when they reach the end of the movement. In particular, the positioning accuracy of the linear axis ensures the precise position control of the workpiece during the processing, thereby ensuring that the size and shape of the processed parts meet the design requirements.
[0025] The dynamic characteristic detection index of the machine tool is to evaluate the degree of dynamic performance degradation of the machine tool by detecting the natural frequency of the machine tool. The natural frequency of a CNC machine tool refers to the vibration frequency generated by itself during operation. In particular, the first-order natural frequency of the machine tool can evaluate the stiffness characteristics of the machine tool at different processing positions, thereby optimizing the machine tool design and improving the overall stiffness of the machine tool; it also helps to avoid resonance during processing and improve processing accuracy.
[0026] Step 2. Establishment of performance degradation dataset
[0027] This patent conducts comprehensive testing on the performance of the entire CNC machine tool during its service life. It uses a laser interferometer to test the change in machine tool accuracy, uses LMS to test the change in dynamic performance of the machine tool, and uses a platinum resistance sensor and a three-axis vibration acceleration sensor to monitor the temperature and vibration data of the machine tool during its service life, and records the actual service time of the machine tool and the working condition parameters of the machine tool during its service life. It can grasp the service performance status of the entire machine under different service times and working conditions of the CNC machine tool, and provide basic data for comprehensive analysis and prediction of the degradation of the machine tool service performance.
[0028] Assume that the data collected includes both intermediate process status data and service performance degradation data under different service times and different working conditions. where s i =[s i1 ,s i2 ,...,s iM ] are parameters of different working conditions, M is the number of working conditions; t i T is the service time; i V is the temperature data of the machine tool during service; i Vibration data of machine tool during service; Positioning accuracy data of machine tool during service;i is the natural frequency data of the machine tool during service; N is the number of data.
[0029] In order to perform cross-validation in machine learning, the order of the data is disrupted by random permutation. According to the needs of the machine tool service performance degradation prediction model based on the nested BP neural network, the collected data is divided into two parts, the training set and the test set. The training set is a data set used to fit the model parameters, and the test set is used to evaluate the performance of the training model. Therefore, the collected data is divided into the training set and the test set in a 1 / 1 ratio.
[0030] Step 3. Establishment of machine tool service performance degradation prediction model
[0031] Step 3.1 Establishment of BP neural network model
[0032] The network model is established by considering the BP model training error, training time, computational efficiency, and regression relationship. The number of input layers is the same as the dimension of the data set, and the hidden layer activation function selects the Sigmoid hyperbolic tangent function; the number of hidden layer neurons is calculated by an empirical formula. In view of the diversity of input data and the applicability of the algorithm, a cyclic calculation method is used in the design to select the optimal number of hidden layer nodes based on the mean square error of the training set.
[0033] The service time and spindle speed parameters of CNC machine tools are used as input vectors. The effective value of the X / Y / Z three-axis vibration velocity of the machine tool spindle and the spindle temperature signal are used as the intermediate process vectors of the performance degradation prediction model. Performance parameters such as positioning accuracy and natural frequency are used as output vectors of the performance degradation prediction model.
[0034] Step 3.2 Establishment of machine tool service performance degradation prediction model
[0035] The performance degradation prediction model based on BP neural network is as follows: Figure 2 As shown, the model consists of two BP neural networks (BP neural network 1 and BP neural network 2), and BP neural network 2 nests BP neural network 1.
[0036] The machine tool service time vector and working condition vector at the past time t are used as input vectors, and the spindle temperature signal and the spindle X / Y / Z three-dimensional vibration signal are used as model output vectors for training; while BP neural network 2 uses the machine tool service time vector, working condition vector, spindle temperature signal, spindle X / Y / Z three-dimensional vibration signal at the past time t as input vectors, and the machine tool positioning accuracy and natural frequency as model output vectors for training.
[0037] The nested BP neural network takes the machine tool service time vector at time t+i, the working condition vector, the spindle temperature result at time t+i generated by BP neural network 1, and the spindle X / Y / Z three-dimensional vibration result as model input to predict the machine tool positioning accuracy and natural frequency at time t+i.
[0038] Step 3.3 Accuracy analysis and prediction of performance degradation prediction model
[0039] BP neural network 1 uses the machine tool service time vector and working condition vector at the past time t as model input, and the spindle temperature signal and spindle X / Y / Z three-axis vibration signal at the past time t as model output for training. 200 sets of data are used to train the performance degradation assessment model, determine the weight of BP neural network 1, and 200 sets of data are used to test the accuracy of BP neural network 1.
[0040] BP neural network 2 takes the machine tool service time vector, working condition vector, spindle temperature signal and spindle X / Y / Z three-dimensional vibration signal at the past time t as model input, and the machine tool positioning accuracy and natural frequency at the past time t as model output for training. 200 groups of data are used to train the performance degradation assessment model, determine the weights of BP neural network 2, and 200 groups of data are used to verify the accuracy of BP neural network 2.
[0041] The nested BP neural network takes the machine tool service time vector at the past time t, the working condition vector, the spindle temperature results at the past time t, and the spindle X / Y / Z three-dimensional vibration results generated by BP neural network 1 as model inputs, and the machine tool positioning accuracy and natural frequency at the past time t as model outputs for training. 200 groups of data are used to train the performance degradation assessment model, determine the weights of the nested BP neural network, and 200 groups of data are used to test the accuracy of the nested BP neural network.
[0042] Taking the machine tool service time vector and working condition vector at the future time t+i as input, the BP neural network 1 predicts the spindle temperature result and the spindle X / Y / Z three-dimensional vibration result at the future time t+i; taking the machine tool service time vector and working condition vector at the future time t+i and the spindle temperature result and the spindle X / Y / Z three-dimensional vibration result at the time t+i predicted based on the BP neural network 1 as input, the nested BP neural network predicts the machine tool positioning accuracy and natural frequency at the future time t+i.
[0043] Example:
[0044] (1) Determination of performance degradation evaluation indicators
[0045] Taking a certain precision vertical machining center as an example, the nested BP neural network is used to analyze the performance degradation prediction results. The performance degradation evaluation index is established based on the accuracy index and dynamic performance index of the CNC machine tool. The accuracy detection index of the machine tool evaluates the performance degradation accuracy of the machine tool by detecting the positioning accuracy of the X-axis of the machine tool; the dynamic performance detection index of the machine tool evaluates the dynamic performance degradation degree of the machine tool by detecting the first-order natural frequency of the machine tool.
[0046] (2) Establishment of performance degradation dataset
[0047] The machining process of a precision vertical machining center was tracked and tested. The cutting parameters of the machining process are shown in Table 1. At the same time, 400 sets of temperature signals, vibration signals, X-axis positioning accuracy, and first-order natural frequency of the whole machine were collected, and the corresponding machine tool startup time and actual cutting parameters were recorded. Among them, 200 sets of experimental data were used as training samples for the performance degradation prediction model, and the remaining 200 sets of data were used as verification samples for the performance degradation prediction model.
[0048] Table 1 Cutting parameters
[0049]
[0050] Table 2 Intermediate process status data and service performance degradation data
[0051]
[0052] (3) Establishment of machine tool service performance degradation prediction model
[0053] 1) Establishment of BP neural network model
[0054] The network model was established by considering the BP model training error, training time, and computational efficiency. The number of training times of the dual neural network was set to 500, the learning rate was set to 0.01, and the target minimum error was set to 0.0001. The number of layers of the input layer of BP neural network 1 was set to 2, and the number of layers of the output layer was set to 4. The number of layers of the input layer of BP neural network 1 was set to 6, and the number of layers of the output layer was set to 2.
[0055] The hidden layer activation function of the dual neural network is the Sigmoid hyperbolic tangent function; the number of neurons in the hidden layer is determined by the empirical formula The calculation results show that n1 is the number of neurons in the hidden layer, n is the number of input layers, a is a constant between 1 and 10, and m is the number of output layers. In view of the diversity of input data and the applicability of the algorithm, a cyclic calculation method is used in the design to select the optimal number of hidden layer nodes based on the mean square error of the training set.
[0056] 2) Establishment of machine tool service performance degradation prediction model
[0057] After running, the network diagram, performance diagram and regression diagram of BP neural network 1 and BP neural network 2 are as follows: Figure 3 As shown in the figure, BP neural network 1 was trained for 20 rounds to reach the target minimum error, and the best validation performance occurred in the 14th round, with a mean square error of 0.000252; the regression plot shows the network output for training, validation, test sets and all targets, the data falls along the 45-degree line, the network output is equal to the target, and in each case, the R value is 0.98 or above.
[0058] BP neural network 2 was trained for 8 rounds to reach the target minimum error. The best verification performance appeared in the 8th round with a mean square error of 1.96×10 -5 ; The regression plot shows the network output for the training, validation, test sets and all targets. Compared with BP neural network 1, the data falls more obviously along the 45-degree line. In each case, the R value is above 0.99.
[0059] 3) Accuracy analysis and prediction of performance degradation prediction model
[0060] The accuracy analysis results of BP neural network 1 are as follows Figure 4 As shown, the spindle temperature accuracy based on the BP neural network 1 is 92.64%, and the spindle X / Y / Z three-axis vibration speed accuracy based on the BP neural network 1 is 97.01% / 97.01% / 97.01% respectively.
[0061] The accuracy analysis results of BP neural network 2 are as follows Figure 4 As shown, the positioning accuracy of the whole machine based on BP neural network 2 is 92.35%, and the accuracy of the first-order natural frequency of the whole machine based on BP neural network 2 is 98.14%.
[0062] The accuracy analysis results of the nested BP neural network are as follows: Figure 4 The accuracy of the whole machine positioning accuracy based on the nested BP neural network is 91.95%, and the accuracy of the first-order natural frequency of the whole machine based on the nested BP neural network is 96.48%.
[0063] Given the condition monitoring data from 0 to 6860 hours, the positioning accuracy and first-order natural frequency degradation results of the whole machine from 6860 to 7000 hours are predicted. The prediction results are as follows: Figure 5 As shown, the prediction result of the positioning accuracy of the 6860-7000h whole machine based on the nested BP neural network is 1.447-1.449μm, and the prediction result of the first-order natural frequency of the 6860-7000h whole machine is 19.675-19.695Hz.
Claims
1. A data-driven nested model-based machine tool service performance degradation prediction method, characterized by: Step 1. Determination of performance degradation evaluation indicators; Establish performance degradation evaluation indicators based on the accuracy indicators and dynamic performance indicators of machine tool performance degradation; the accuracy detection indicators of machine tools evaluate the performance degradation accuracy of machine tools by detecting the positioning accuracy of machine tools, etc.; the dynamic characteristics detection indicators of machine tools evaluate the degree of dynamic performance degradation of machine tools by detecting the natural frequency of machine tools, etc.; Step 2. Establishment of performance degradation dataset; A performance degradation data set is established by collecting intermediate process status data and service performance degradation data under different service times and different working conditions. i=1,2,...,N,where s i =[s i1 ,s i2 ,...,s iM ] are parameters of different working conditions, M is the number of working conditions; t i T is the service time; i V is the temperature data of the machine tool during service; i Vibration data of machine tool during service; Positioning accuracy data of machine tool during service; i is the natural frequency data of the machine tool service process; N is the number of data; Step 3. Establishment of machine tool service performance degradation prediction model; The service time and spindle speed parameters of the CNC machine tool are used as input vectors; the effective value of the X / Y / Z three-dimensional vibration velocity of the machine tool spindle and the spindle temperature signal are used as the intermediate process vectors of the performance degradation prediction model; the performance parameters such as positioning accuracy and natural frequency are used as the output vectors of the performance degradation prediction model; A performance degradation prediction model based on a nested structure is established. The model consists of two degradation models, including degradation model 1 and degradation model 2. Degradation model 2 nests degradation model 1. In order to predict the service performance prediction results of the machine tool at the future time t+i, the machine tool service time vector and working condition vector at the future time t+i are set as input, and the spindle temperature results and spindle X / Y / Z three-axis vibration results at the future time t+i are predicted through degradation model 1; the machine tool service time vector and working condition vector at the time t+i and the spindle temperature results and spindle X / Y / Z three-axis vibration results predicted based on degradation model 1 are taken as input, and the machine tool positioning accuracy results and natural frequency results at the future time t+i are predicted through degradation model 2.
2. The data-driven nested model-based machine tool service performance degradation prediction method according to claim 1, characterized in that: The implementation process of the machine tool service performance degradation prediction model is as follows: Step 3.1 Establishment of BP neural network model; The network model is established by considering the BP model training error, training time, computational efficiency, and regression relationship; the number of input layers is the same as the dimension of the data set, and the hidden layer activation function selects the Sigmoid hyperbolic tangent function; the number of hidden layer neurons is calculated by an empirical formula; in view of the diversity of input data and the applicability of the algorithm, a cyclic calculation method is used in the design to select the optimal number of hidden layer nodes based on the mean square error of the training set; The service time and spindle speed parameters of the CNC machine tool are used as input vectors; the effective value of the X / Y / Z three-dimensional vibration velocity of the machine tool spindle and the spindle temperature signal are used as the intermediate process vectors of the performance degradation prediction model; the performance parameters such as positioning accuracy and natural frequency are used as the output vectors of the performance degradation prediction model; Step 3.2 Establishment of machine tool service performance degradation prediction model; The performance degradation prediction model established based on the BP neural network is composed of two BP neural networks (BP neural network 1 and BP neural network 2), and the BP neural network 2 nests the BP neural network 1; The machine tool service time vector and working condition vector at the past time t are used as input vectors, and the spindle temperature signal and the spindle X / Y / Z three-dimensional vibration signal are used as model output vectors for training; while the BP neural network 2 uses the machine tool service time vector, working condition vector, spindle temperature signal, spindle X / Y / Z three-dimensional vibration signal at the past time t as input vectors, and the machine tool positioning accuracy and natural frequency as model output vectors for training; The nested BP neural network takes the machine tool service time vector at time t+i, the working condition vector, the spindle temperature result at time t+i generated by BP neural network 1, and the spindle X / Y / Z three-dimensional vibration result as model input to predict the machine tool positioning accuracy and natural frequency at time t+i; Step 3.3 Accuracy analysis and prediction of performance degradation prediction model; BP neural network 1 uses the machine tool service time vector and working condition vector at the past time t as model input, and the spindle temperature signal and spindle X / Y / Z three-dimensional vibration signal at the past time t as model output for training; 200 sets of data are used to train the performance degradation assessment model, determine the weights of BP neural network 1, and 200 sets of data are used to test the accuracy of BP neural network 1; BP neural network 2 takes the machine tool service time vector, working condition vector, spindle temperature signal and spindle X / Y / Z three-dimensional vibration signal at the past time t as model input, and the machine tool positioning accuracy and natural frequency at the past time t as model output for training. 200 sets of data are used to train the performance degradation assessment model, determine the weight of BP neural network 2, and 200 sets of data are used to test the accuracy of BP neural network 2. The nested BP neural network takes the machine tool service time vector at the past time t, the working condition vector, the spindle temperature results at the past time t, and the spindle X / Y / Z three-dimensional vibration results generated by BP neural network 1 as model inputs, and the machine tool positioning accuracy and natural frequency at the past time t as model outputs for training. 200 groups of data are used to train the performance degradation assessment model, determine the weights of the nested BP neural network, and 200 groups of data are used to test the accuracy of the nested BP neural network.
3. The data-driven nested model-based machine tool service performance degradation prediction method according to claim 2, characterized in that: Taking the machine tool service time vector and working condition vector at the future time t+i as input, the BP neural network 1 predicts the spindle temperature result and the spindle X / Y / Z three-dimensional vibration result at the future time t+i; taking the machine tool service time vector and working condition vector at the future time t+i and the spindle temperature result and the spindle X / Y / Z three-dimensional vibration result at the time t+i predicted based on the BP neural network 1 as input, the nested BP neural network predicts the machine tool positioning accuracy and natural frequency at the future time t+i.