Machine tool energy consumption prediction method and system based on integrated model and feature driving

Abstract

The invention discloses a machine tool energy consumption prediction method based on an integrated model and feature driving, and the method comprises the steps: obtaining the machining data of different machine tools, and dividing the machining data of different machine tools into a sample set of a source domain and a sample set of a target domain; preprocessing the sample sets of the source domain and the target domain, and extracting feature vectors; representing an energy consumption dynamic change rule of the machine tool in the source domain and the target domain based on a Gaussian process, carrying out correlation analysis on feature vectors of the source domain and the target domain, and adopting an isomorphic feature migration method by utilizing correlation; using the feature vector based on the source domain as a training set to train a Boosting model as an energy consumption prediction model; and performing regression prediction on the energy consumption of the target domain based on the trained energy consumption prediction model, and performing stage decomposition on the predicted energy consumption based on a Gaussian process to obtain a predicted value of each stage of machine tool processing. The problem of energy consumption prediction difference of different machine tools is solved through feature migration in combination with a data-mechanism model, and the method can be applied to the field of energy consumption prediction of universal machine tool machining.

Description

Technical Field

[0001] The present invention relates to energy consumption prediction of digital manufacturing equipment, and in particular to a method and system for predicting energy consumption of machine tools based on integrated model and feature drive. Background Art

[0002] Effective lean control of energy consumption provides technical support for energy-saving manufacturing of CNC equipment. However, in the actual processing process, the differences in feature distribution caused by different scenarios and different processing parameters seriously reduce the accuracy of prediction. As the most commonly used process in digital manufacturing, CNC machine tool processing can flexibly complete product processing tasks with high surface quality and complex structure, but it is accompanied by huge energy consumption, which increases the negative impact on the environment. The existing technology attempts to accurately predict the energy consumption of CNC to facilitate decision-making on process routes, process parameters, etc. However, due to interference factors from multiple sources and expensive data acquisition systems, it makes the prediction of processing energy consumption under different working conditions a challenge.

[0003] In addition, existing machine learning methods rely on a large number of labeled samples with the same probability distribution, which leads to a significant decrease in the performance of traditional prediction methods. To this end, based on the multiple disturbance factors of energy consumption, the energy consumption is quantitatively analyzed based on the existing physical information, so as to predict the energy consumption by fitting the quantitative relationship. However, the quantitative model lacks certainty and robustness when solving the problem of limited samples and coupling of multiple physical quantities. Summary of the invention

[0004] Purpose of the invention: In view of the above problems, the present invention provides a machine tool energy consumption prediction method and system based on integrated model and feature drive which has universality and high prediction accuracy.

[0005] Technical solution: To solve the above problems, the present invention adopts a machine tool energy consumption prediction method based on integrated model and feature drive, which includes the following steps:

[0006] (1) Obtaining processing data of different machine tools and dividing the processing data of different machine tools into a sample set of the source domain and a sample set of the target domain;

[0007] (2) Preprocess the sample sets of the source domain and the target domain and extract feature vectors;

[0008] (3) Based on the Gaussian process, the dynamic change law of energy consumption of machine tools in the source domain and the target domain is characterized, and the correlation analysis of the feature vectors of the source domain and the target domain is performed. The isomorphic feature migration method is used based on the correlation;

[0009] (4) Using the feature vector based on the source domain as the training set to train the Boosting model as an energy consumption prediction model;

[0010] (5) Based on the trained energy consumption prediction model, the target domain energy consumption is regressed and predicted, and the predicted energy consumption is decomposed into stages based on the Gaussian process to obtain the predicted values ​​for each stage of machine tool processing.

[0011] Furthermore, in step (3), the dynamic change law of energy consumption of machine tools in the source domain and the target domain is characterized based on Gaussian process, including feature-based stage energy consumption decomposition in the source domain and time scale decomposition in the target domain. The specific expression is:

[0012]

[0013]

[0014] Among them, EC s represents the total energy consumption of the source domain, EC t represents the total energy consumption of the target domain; t s and t e Indicates the start and end time of processing; P s (t) represents the power of the machine tool during standby phase, P ta (t) represents the power of the machine tool during the startup phase; C represents a constant determined by the spindle's rotational inertia; n represents the machine tool's rotational speed, v represents the machine tool's feed speed, and a p Indicates the cutting depth of the machine tool, a e represents the cutting width of the machine tool; f(·) and g(·) represent multi-source nonlinear functions; VB represents the wear of the flank; χ represents the state function of the cooling system; k represents the total number of tool passes in a single machining operation of the machine tool; P c represents the average cutting power; μ represents the time scale in the Gaussian distribution.

[0015] Furthermore, in step (3), the correlation analysis of the feature vectors of the source domain and the target domain is performed using the bulldozer distance representation, and the bulldozer distance is expressed as:

[0016]

[0017] Among them, Γ(F1, F2) represents all possible joint distributions of the feature space F1 composed of the source domain feature vectors and the feature space F2 composed of the target domain feature vectors; z1 and z2 represent the feature values ​​of the same features in the source domain and the target domain; p represents the total number of features in the feature space, Represents a set of integers.

[0018] Furthermore, the preprocessing of the sample sets in the source domain and the target domain in step (2) includes using a Kalman filter to perform noise reduction and smoothing on the original power signal.

[0019] Furthermore, the feature extraction algorithm used in extracting the feature vector in step (2) is a recurrent neural network (RNN).

[0020] Furthermore, in the step (4), the AdaBoost integrated model in the Boosting model is selected as the energy consumption prediction model, and the training set obtains the optimal hyperparameter combination through Monte Carlo cross-validation. The Monte Carlo cross-validation enables the learner to train the training set through feature selection and weight allocation of the decision tree, and the decision tree model is optimized by modifying the number of decision trees and adjusting the maximum depth method.

[0021] The present invention also adopts a machine tool energy consumption prediction system based on integrated model and feature drive, comprising:

[0022] A data acquisition module is used to acquire processing data of different machine tools and divide the processing data of different machine tools into a sample set of a source domain and a sample set of a target domain;

[0023] The feature extraction module is used to preprocess the sample sets of the source domain and the target domain and extract feature vectors;

[0024] The association module is used to characterize the dynamic change law of energy consumption of machine tools in the source domain and the target domain based on Gaussian process, perform correlation analysis on the feature vectors of the source domain and the target domain, and use the isomorphic feature migration method based on the correlation;

[0025] A model building module is used to train a Boosting model as an energy consumption prediction model by using a feature vector based on a source domain as a training set;

[0026] The prediction module is used to perform regression prediction on the target domain energy consumption based on the trained energy consumption prediction model, and decompose the predicted energy consumption into stages based on the Gaussian process to obtain the predicted values ​​of each stage of machine tool processing.

[0027] Furthermore, the data acquisition module includes a tool health status monitoring system, a physical signal acquisition system, an electrical signal acquisition system and a network communication system, and the data acquisition module includes hardware devices: an optical microscope, a three-dimensional dynamometer, a three-dimensional vibration dynamometer, a charge amplifier, a multi-function meter and an edge gateway. The present invention also uses a computer device, including a memory, a processor and a computer program stored in the memory and executable on the processor, and the processor implements the steps of the above method when executing the computer program.

[0028] The present invention also adopts a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the above method are implemented.

[0029] Beneficial effects: Compared with the prior art, the significant advantage of the present invention is that it solves the problem of energy consumption prediction differences of different machine tools by combining feature migration with data-mechanism model, and can be applied to the field of energy consumption prediction of general machine tool processing. It overcomes the limitation that energy consumption prediction is not universal in terms of sample quantity and sample feature distribution differences, and improves the accuracy of energy consumption prediction of machine tool processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a schematic diagram of the flow chart of the energy consumption prediction method of the present invention;

[0031] Figure 2 It is a structural schematic diagram of the energy consumption prediction system of the present invention. DETAILED DESCRIPTION

[0032] Depend on Figure 1 As shown, in this embodiment, a machine tool energy consumption prediction method based on integrated model and feature drive includes the following steps:

[0033] (1) The energy consumption control platform is used to control the data acquisition module to obtain the processing data of different machine tools through the real-time integration method, and the processing data of different machine tools are divided into energy consumption sample sets of source domain and target domain scenarios. The data acquisition module includes a tool health status monitoring system, a physical signal acquisition system, an electrical signal acquisition system, and a network communication system. The hardware equipment of the data acquisition module includes an optical microscope, a three-axis dynamometer, a three-axis vibration dynamometer, a charge amplifier, a multi-function meter, and an edge gateway, such as Figure 2 shown.

[0034] The experimental parameter groups were combined using a four-factor three-level orthogonal design method. The cooling method and milling method of the source domain scene were mixed by dry cutting and wet cutting, and face milling and end milling to ensure the diversity of the original input features. The processing parameters of the source domain and the target domain are shown in Table 1.

[0035] Table 1 Experimental parameter groups

[0036] Scenario Cooling method Milling method n(rpm) v(mm / min) <![CDATA[a p (mm)]]> <![CDATA[a e (mm)]]> source domain Dry cutting / wet cutting Face milling / end milling [700,900] [100,200] [0.2,0.6] [5,20] Target domain Dry Cut Face milling [5000,7000] [300,420] [0.25,1.4] [0.2,0.5]

[0037] Among them, the source domain scenario is based on the CY-VMC850Le CNC machine tool for processing, and the power data acquisition uses the WB9128 power sensor to communicate locally with the PLC controller; the target domain scenario is based on the DX-650 engraving and milling machine for processing, and the signal acquisition is synchronized with the multi-function meter through the DEWESoft signal acquisition device, and the local and network dual communication is realized through the human-machine interface and the edge gateway.

[0038] (2) Preprocess the sample sets of the source domain and the target domain, and extract the feature vector. The Kalman filter is used to reduce noise and smooth the original power signal, and the feature vector is calculated using the feature extraction algorithm. The feature extraction algorithm is mainly a recurrent neural network (RNN). Based on the time series power signal, the RNN gating unit is used to capture the signal features to form a feature space, which is used as the input of the prediction model.

[0039] (3) The Gaussian process is used to characterize the dynamic change law of the energy consumption of machine tools in the source domain and the target domain. The correlation between the feature vectors of the source domain and the target domain is analyzed. The energy consumption of the target domain is predicted by using the instance-based isomorphic feature transfer method based on the correlation, and the feature correlation is evaluated based on the bulldozer distance.

[0040] The energy consumption mapping relationship is completed by using the feature-based stage energy consumption decomposition of the source domain and the time scale decomposition of the target domain, and the correlation analysis of the scene is performed based on the energy consumption characteristics. Therefore, the energy consumption model of the source domain and the target domain is as follows:

[0041]

[0042] Among them, EC s With EC t are the total energy consumption of the source domain and the target domain respectively; t s and t e Represents the start and end time of processing; P s (t) represents the power of the machine tool during the standby phase, usually a constant value; P ta (t) is the power in the startup phase, which is very short and is usually 6 times the standby power. C is a constant determined by the spindle's rotational inertia. n, v, a p , a e are the four elements of machine tool cutting, namely rotation speed, feed speed, cutting depth and cutting width; f(·) and g(·) represent multi-source nonlinear functions; VB and χ are the flank wear and cooling system state functions, respectively; k is the total number of tool passes in a single machining operation of the machine tool; P c is the average cutting power, and μ is the time scale in the Gaussian distribution.

[0043] The bulldozer distance can measure the shortest distance under the same feature dimension, and can also be understood as the difference in scenes. In the field of machine tool processing, the bulldozer distance in different scenes can be expressed as:

[0044]

[0045] Among them, Γ(F1, F2) represents all possible joint distributions of the feature space F1 composed of the source domain feature vectors and the feature space F2 composed of the target domain feature vectors; z1 and z2 represent the feature values ​​of the same features in the source domain and the target domain; p represents the total number of features in the feature space, Represents an integer set. Based on the scenario where the source domain and the target domain are homologous and isomorphic, this paper uses 12 features of RNN, and the calculation results are shown in Table 2:

[0046] Table 2. Results of feature correlation analysis

[0047]

[0048] It can be seen from Table 2 that the bulldozer distance calculated through the feature space of the source domain and the target domain is less than 1, which meets the condition of feature migration. At the same time, it proves that the source domain and the target domain are more closely related in the homologous and isomorphic scenario, and the rationality and reliability of feature migration are higher.

[0049] (4) The feature vector based on the source domain is used as the training set to train the Boosting model as an energy consumption prediction model. The AdaBoost ensemble model is selected as the energy consumption prediction model, and the data set is used to obtain the optimal hyperparameter combination through Monte Carlo cross-validation. After Monte Carlo cross-validation, AdaBoost enables the learner to train the training set through feature selection and weight allocation of the decision tree, optimize the decision tree model by modifying the number of decision trees, adjusting the maximum depth, and predicting the energy consumption of the target domain.

[0050] (5) Based on the trained energy consumption prediction model, the target domain energy consumption is regressed and predicted, and the target domain energy consumption is generally predicted based on AdaBoost. The energy consumption of each stage is compared based on the time scale of the Gaussian model.

[0051] Six groups of target domain experiments are selected to demonstrate the energy consumption and prediction effects. The processing parameters and energy consumption prediction are shown in Table 3.

[0052] Table 3 Target domain experimental parameters and energy consumption

[0053] Exp.no <![CDATA[(n,v,a p ,a e )]]> Actual energy consumption (Wh) Predicted energy consumption (Wh) 1 (5000,300,0.25,0.2) 3.118 3.152 2 (5000,360,0.6,0.5) 2.847 2.854 3 (6000,420,0.25,0.35) 2.647 2.636 4 (6000,360,0.6,0.35) 3.296 3.284 5 (7000,360,0.25,0.35) 3.315 3.317 6 (7000,420,0.25,0.5) 3.208 3.227

[0054] It can be seen from Table 3 that the overall feature space of the source domain is migrated, and the prediction error of the energy consumption value of the target domain is less than 10%, indicating that the feature correlation between the source domain and the target domain after RNN feature extraction is strong, which preliminarily illustrates the effectiveness and feasibility of energy consumption prediction based on feature migration.

[0055] The energy consumption is decomposed into stages based on the Gaussian process model and compared with the actual energy consumption, as shown in Table 4.

[0056] Table 4 Stage energy consumption and error analysis

[0057]

[0058] As shown in Table 4, the change patterns of the predicted values ​​and measured values ​​of the stage energy consumption distribution are very similar, which verifies the feasibility of using the Gaussian process model to model the general milling energy consumption. In addition, the energy consumption in the standby stage has almost no change, and the change patterns of the startup stage, the no-load stage and the feed stage are similar, which shows that the changes in process parameters have a great consistency in the impact of these three stages. The prediction accuracy of the overall energy consumption and the stage energy consumption is above 95%, which proves that this method can achieve excellent prediction results in different milling scenarios, and can be extended to various general CNC machining processes.

Claims

1. A machine tool energy consumption prediction method based on integrated model and feature drive, characterized in that: The following steps are involved: (1) Obtaining processing data of different machine tools and dividing the processing data of different machine tools into a sample set of the source domain and a sample set of the target domain; (2) Preprocess the sample sets of the source domain and the target domain and extract feature vectors; (3) Based on the Gaussian process, the dynamic change law of energy consumption of machine tools in the source domain and the target domain is characterized, and the correlation analysis of the feature vectors of the source domain and the target domain is performed. The isomorphic feature migration method is used based on the correlation; (4) Using the feature vector based on the source domain as the training set to train the Boosting model as an energy consumption prediction model; (5) Based on the trained energy consumption prediction model, the target domain energy consumption is regressed and predicted, and the predicted energy consumption is decomposed into stages based on the Gaussian process to obtain the predicted values ​​for each stage of machine tool processing.

2. The machine tool energy consumption prediction method according to claim 1, characterized in that: In step (3), the dynamic change law of energy consumption of machine tools in the source domain and the target domain is characterized based on Gaussian process, including feature-based stage energy consumption decomposition in the source domain and time scale decomposition in the target domain. The specific expression is: Among them, EC s represents the total energy consumption of the source domain, EC t represents the total energy consumption of the target domain; t s and t e Indicates the start and end time of processing; P s (t) represents the power of the machine tool during standby phase, P ta (t) represents the power of the machine tool during the startup phase; C represents a constant determined by the spindle's rotational inertia; n represents the machine tool's rotational speed, v represents the machine tool's feed speed, and a p Indicates the cutting depth of the machine tool, a e represents the cutting width of the machine tool; f(·) and g(·) represent multi-source nonlinear functions; VB represents the wear of the flank; χ represents the state function of the cooling system; k represents the total number of tool passes in a single machining operation of the machine tool; P c represents the average cutting power; μ represents the time scale in the Gaussian distribution.

3. The machine tool energy consumption prediction method according to claim 2, characterized in that: In step (3), the correlation analysis of the feature vectors of the source domain and the target domain is performed using the bulldozer distance representation, which is expressed as: Among them, Γ(F1, F2) represents all possible joint distributions of the feature space F1 composed of the source domain feature vectors and the feature space F2 composed of the target domain feature vectors; z1 and z2 represent the feature values ​​of the same features in the source domain and the target domain; p represents the total number of features in the feature space, Represents a set of integers.

4. The machine tool energy consumption prediction method according to claim 2, characterized in that: The preprocessing of the sample sets of the source domain and the target domain in the step (2) includes using a Kalman filter to perform noise reduction and smoothing processing on the original power signal.

5. The machine tool energy consumption prediction method according to claim 3, characterized in that: The feature extraction algorithm used to extract the feature vector in step (2) is a recurrent neural network (RNN).

6. The machine tool energy consumption prediction method according to claim 5, characterized in that: In the step (4), the AdaBoost integrated model in the Boosting model is selected as the energy consumption prediction model, and the training set obtains the optimal hyperparameter combination through Monte Carlo cross validation. The Monte Carlo cross validation enables the learner to train the training set through feature selection and weight allocation of the decision tree, and the decision tree model is optimized by modifying the number of decision trees and adjusting the maximum depth method.

7. A machine tool energy consumption prediction system based on integrated model and feature drive, characterized in that: include: A data acquisition module is used to acquire processing data of different machine tools and divide the processing data of different machine tools into a sample set of a source domain and a sample set of a target domain; The feature extraction module is used to preprocess the sample sets of the source domain and the target domain and extract feature vectors; The association module is used to characterize the dynamic change law of energy consumption of machine tools in the source domain and the target domain based on Gaussian process, perform correlation analysis on the feature vectors of the source domain and the target domain, and use the isomorphic feature migration method based on the correlation; A model building module is used to train a Boosting model as an energy consumption prediction model by using a feature vector based on a source domain as a training set; The prediction module is used to perform regression prediction on the target domain energy consumption based on the trained energy consumption prediction model, and decompose the predicted energy consumption into stages based on the Gaussian process to obtain the predicted values ​​of each stage of machine tool processing.

8. The machine tool energy consumption prediction system according to claim 7, characterized in that: The data acquisition module includes a tool health status monitoring system, a physical signal acquisition system, an electrical signal acquisition system and a network communication system. The data acquisition module includes hardware equipment: an optical microscope, a three-axis dynamometer, a three-axis vibration dynamometer, a charge amplifier, a multi-function meter and an edge gateway.

9. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images