High-speed machining anomaly detection method based on probability optimization support vector data description

Abstract

The invention discloses a high-speed machining anomaly detection method based on probability optimization support vector data description. The high-speed machining anomaly detection method comprises the steps of 1, collecting vibration signals in a high-speed machining process in real time; step 2, performing feature extraction on the collected vibration signals based on AI; step 3, randomly classifying the extracted feature samples; step 4, probability estimation is carried out; 5, an SVDD model is constructed through Ftrain; step 6, performing weighted optimization; step 7, obtaining expressions of the model center O and the decision boundary R; and 8, integrating the SVDD model subjected to probability density regularization obtained in the step 7 into a high-speed processing machine tool monitoring platform based on a friend machine technology, and monitoring abnormal conditions in the processing process in real time. The method does not depend on a large number of balance samples, probability estimation is carried out on feature samples in advance, on one hand, the detection precision is improved, on the other hand, the false alarm rate is reduced, the optimized SVDD model is integrated to a high-speed machining process monitoring platform, and a high-speed machining machine tool is protected; the method has a good application prospect.

Description

Technical Field

[0001] The present invention relates to the field of unsupervised anomaly detection methods for high-speed machining processes, and in particular to a high-speed machining anomaly detection method based on probability optimization support vector data description. Background Art

[0002] With the development of science and technology, manufacturing has become an important driving force for economic development. As the core equipment in the manufacturing industry, CNC machine tools have been widely used in aerospace, energy and chemical, and military industries, and are gradually developing in the direction of high precision, ultra-high speed and intelligence. Among them, high-speed machining of machine tools is an advanced manufacturing technology with the fastest research progress, the widest application field and the most significant economic benefits. High-speed machining can cut, mill, drill and other operations on workpieces through high-speed rotating tools in a short time, and process parts quickly and efficiently. High-speed machining has the advantages of high speed, high efficiency and high precision, and has been widely used in high-precision fields such as military equipment and aerospace. However, due to the contact, friction, vibration and other factors between the tool and the workpiece during high-speed machining, as well as the influence of factors such as the non-uniformity of the material itself, some abnormal conditions may occur during the machining process. These abnormalities include broken tool blades, chip jamming of the tool holder, spindle bearing failure and other problems. These abnormal conditions will not only reduce product quality, but also affect production efficiency and personal safety. In order to improve the quality and efficiency of the machining process and reduce the occurrence of abnormal conditions, it is necessary to perform abnormal detection on the high-speed machining process. In the prior art, high-speed machining anomaly detection is to monitor in real time whether there are any abnormal conditions in the machining process by monitoring and analyzing the monitoring signals during the machining process, such as cutting force, vibration signal, power signal, etc. By timely discovering abnormal machining conditions, corresponding measures can be taken to adjust and repair them to improve the stability and reliability of the high-speed machining process. However, in many of the above cases, it is necessary to manually combine the data collected by various sensors for judgment, so the efficiency is not high, and it is affected by the skills and experience of technicians, and the judged data often cannot truly reflect the specific situation.

[0003] In recent years, with the continuous development of sensor technology and industrial Internet of Things, as well as the application of artificial intelligence and machine learning algorithms, the field of high-speed machining anomaly detection has made significant progress. By real-time monitoring and analyzing the monitoring data of the machining process, the abnormal conditions of the high-speed machining process can be identified and warned. In particular, intelligent algorithms represented by deep learning have made great achievements in the field of intelligent manufacturing. The first method is to use hybrid robust convolutional autoencoders for tool anomaly detection under noise interference. This method constructs a novel deep learning framework for tool wear anomaly detection in ultra-precision machining; the second method is that a probabilistic model based on digital-analog linkage is proposed for the stability analysis method of the grinding process. The above two mainstream data-driven anomaly detection methods can improve the automation and intelligence level of machining, and provide more reliable and efficient machining solutions for "intelligent manufacturing". Although the above methods have achieved excellent results in the field of machining process state monitoring, there are still many challenges for anomaly detection in high-speed machining in the face of complex unlabeled data and serious imbalanced samples. First, unlike laboratory data, most of the data collected in actual industrial sites lack accurate labels, which makes many supervised methods ineffective; second, the high-speed machining process is in normal processing conditions most of the time, which leads to an extreme lack of abnormal samples. The positive and negative samples of the collected data are extremely unbalanced, and deep learning methods that overly rely on balanced samples are often difficult to use for anomaly detection; finally, the collected data inevitably contains various noise interferences, which brings challenges to data-driven anomaly detection methods. For the above reasons, high-speed machining anomaly detection based on artificial intelligence and machine learning algorithms still has certain shortcomings.

[0004] As an unsupervised (unsupervised is a machine learning training method, that is, machine learning is performed without labels, and its corresponding methods are "supervised" and "semi-supervised") anomaly detection method, Support Vector Data Description (SVDD) is widely used in machining process status monitoring. It does not require too many abnormal samples, and even only normal samples can be used to monitor the health of high-speed machining processes. In addition, compared with deep learning similar to black box models, SVDD can provide an intuitive description of the detection results. With these advantages, SVDD is used in many intelligent monitoring systems in production to improve the safety and efficiency of the machining process. However, the existing SVDD method lacks the ability to detect abnormal data that is confused in normal samples, which will lead to a decrease in the detection accuracy of SVDD. Specifically, when applied to abnormality detection in high-speed machining of machine tools, since the high-speed machining process of machine tools is in normal processing state most of the time, the collected monitoring data will be seriously unbalanced in both quantity and quality. Especially for small-batch processed products, there is not even abnormal data for model training, which makes the traditional supervised learning method invalid. In addition, the monitoring data collected at the industrial site will inevitably be interfered by various noises. Even in normal samples, some abnormal data will inevitably be confused. These abnormal data will interfere with the SVDD model.

[0005] Therefore, unsupervised anomaly detection methods cannot be effectively applied to anomaly detection in high-speed machining of machine tools. Summary of the invention

[0006] In order to overcome the disadvantage that the existing unsupervised anomaly detection methods cannot be effectively applied to anomaly detection in high-speed machining of machine tools due to technical limitations, the present invention provides a method that can quickly and effectively evaluate the reliability of high-speed machining without interrupting the machining process under the joint action of related processes, so as to improve the machining accuracy and efficiency. In addition, in order to avoid the interference of abnormal data on the SVDD model, the probability density of the data is estimated, and the SVDD is improved using the estimated probability to enhance the model's description ability of noisy data, thereby improving the high-speed machining anomaly detection method based on probability optimized support vector data description, which improves the anomaly detection accuracy in high-speed machining of machine tools.

[0007] The technical solution adopted by the present invention to solve its technical problem is:

[0008] The high-speed machining anomaly detection method based on probability optimization support vector data description is characterized in that it includes the following steps: step 1, using vibration sensors installed on the machine tool spindle and the workbench to collect vibration signals in the high-speed machining process in real time; step 2, extracting features from the collected vibration signals to suppress noise interference on the one hand and improve detection efficiency on the other hand. The extracted features include the root mean square value M in the time domain and the spectrum energy P in the frequency domain, which are recorded as the feature sample set F=[F1,F2,...,F K ] T , where each feature sample is F k ={M k ,P k}; Step 3: Randomly classify the extracted feature samples and establish training sets F train and the test set F test ; Step 4: Use the Gaussian kernel function to train the training set F constructed in step 3 train Probability estimation is performed, and the estimated probability of each sample data is recorded as p(F i ); Step 5, through the training set F train Construct an SVDD model, with the model center denoted as O and the decision boundary denoted as R. The constructed SVDD does not require abnormal samples, and only normal samples are needed to construct a decision boundary for determining whether the high-speed machining process is abnormal. Step six, use the probability density estimated in step four to perform weighted optimization on the SVDD model constructed in step five, and assign weights to each sample based on the probability density. Step seven, use the Lagrange multiplier method to solve the model established in step six, and obtain the expressions of the model center O and the decision boundary R. Input the test set constructed in step three into the optimized SVDD model to identify the abnormal conditions of each test sample. Step eight, integrate the probability density regularized SVDD model obtained in step seven into the high-speed machining machine tool monitoring platform of "Friend Machine Technology", so that the monitoring system can communicate and interact with the machine tool in real time, and monitor abnormal conditions in the machining process in real time.

[0009] Furthermore, in step 1, the vibration signal formula is D={(x i ,y i ,z i ),(x j ,y j ,z j )},i,j=1,2,...,N.

[0010] Furthermore, in step three, the training set is used to construct the SVDD model, and the test set is used to verify the anomaly detection accuracy of the model.

[0011] Furthermore, in step 4, the sample probability density at the center of the data feature distribution is large, and the sample probability density at the edge of the data feature distribution is small, and the estimated probability density indicates the importance of each sample.

[0012] Furthermore, in step five, if there are a small number of abnormal samples, the data description constructed by SVDD can be optimized using the abnormal samples to improve the anomaly detection accuracy of SVDD.

[0013] Furthermore, in step six, the allocation strategy is not only explainable, but also can adaptively determine the influence of each sample on the SVDD model according to the distribution of data samples. The optimized model effectively improves the detection accuracy.

[0014] Furthermore, in step seven, if the distance between the test sample and the model center exceeds the decision radius R, it is marked as abnormal, otherwise it is normal, and the accuracy of the optimization model is statistically calculated.

[0015] Furthermore, in step eight, once an abnormality is detected, the system can promptly issue an alarm or provide corresponding feedback to ensure the safe and stable operation of the high-speed machining machine.

[0016] The beneficial effects of the present invention are as follows: based on the existing SVDD unsupervised anomaly detection method, the present invention can quickly and effectively evaluate the reliability of high-speed machining without interrupting the machining process under the joint action of related processes, so as to improve the machining accuracy and efficiency; specifically, the difficulties faced by the actual machining process are fully considered, and only a small number of normal samples are needed to construct the decision boundary of anomaly detection without relying on a large number of balanced samples; the probability of feature samples is estimated in advance, and the noise interference confused in the normal samples can be effectively avoided according to the estimated probability, thereby improving the accuracy and generalization ability of anomaly detection, so that the constructed anomaly detection method can adapt to actual industrial scenarios; the SVDD model is optimized by the estimated probability, which effectively enhances the description ability of the SVDD model for complex data distribution, improves the detection accuracy on the one hand, and reduces the false alarm rate on the other hand, and the optimized SVDD model is integrated into the high-speed machining process monitoring platform to escort high-speed machining machine tools; based on the above, the present invention has good application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a schematic diagram of the structure of the high-speed machining machine tool of the present invention.

[0018] Figure 2 This is a schematic diagram of the installation of the vibration sensor of the present invention.

[0019] Figure 3 It is a schematic diagram of abnormal tool chipping during high-speed machining of the present invention.

[0020] Figure 4 It is a process framework diagram of the present invention.

[0021] Figure 5 This is the probability density distribution diagram of the training samples of the present invention.

[0022] Figure 6 This is the traditional SVDD model diagram of the present invention.

[0023] Figure 7 This is the SVDD model diagram of probability density regularization of the present invention.

[0024] Figure 8 This is the confusion matrix diagram of high-speed machining anomaly detection of the present invention. DETAILED DESCRIPTION

[0025] Figure 1 As shown, the present invention takes a small data processing center (FANUCα-T14iFb) as an example for explanation, and the machine tool structure diagram is shown in Figure 1 As shown, the machine tool spindle speed is 20000rpm, the workpiece material is aluminum alloy, and the installation position of the vibration sensor (IEPE8688A5) used to collect monitoring data is as follows Figure 2 As shown, the sampling frequency is 6250Hz, which is used to test the abnormality of the tool edge breaking in the present invention. The schematic diagram of the abnormal tool is as shown in Figure 3 As shown, the material is cemented carbide (WC-TiC-TaC-Co).

[0026] Figure 4 As shown in FIG, the high-speed machining anomaly detection method based on probability optimization support vector data description has the following specific steps: Step 1, as shown in FIG. Figure 1 As shown in the figure, vibration sensors installed on the tool spindle and workbench of the machine tool are used to collect vibration signals in the high-speed machining process in real time. The sampling frequency is 6250 Hz, the sampling time is 1 hour, and the sampling channels are 3 (x, y, z). The collected signal is recorded as D = {(x i ,y i ,z i ),(x j ,y j ,z j )},i,j=1,2,...,N, where i represents the spindle vibration signal number, j represents the workbench vibration signal number. This step provides monitoring data for the feature extraction of step 2. Step 2, extract the features of the collected vibration signal according to formula (1). In order to construct the feature sample set, it is calculated once every 1s. In formula (1), each 1s contains 6250 data points, so T=6250, a total of 3600 feature samples, so K=3600, where x i Represents the signal of each channel, f irepresents the spectrum of the corresponding channel signal, and the constructed feature sample set is recorded as F = [F1, F2, ..., F K ] T , where each feature sample is F k ={M k ,P k}; This step provides feature samples for data partitioning in step three;

[0027] Figure 4 As shown, in step 3, the feature sample set F constructed in step 2 is divided, where 80% of the data samples are used for training to construct the training set F train , the remaining 20% ​​of the data samples are used for testing and construct the test set F test , this step provides training samples and test samples for step 4; step 4, according to formula (2), use the Gaussian kernel function to train the training set F constructed in step 3 train Probability estimation is performed, and the estimated probability of each sample is recorded as p(F i ),like Figure 5 As shown in the figure, it can be seen that the sample probability density at the center of the data feature distribution is large, and the sample probability density at the edge of the data feature distribution is small. The estimated probability density indicates the importance of each sample; Where k() represents the Gaussian kernel function, σ represents the kernel function parameter, and this step provides training samples for model construction in step 5;

[0028] Figure 4 As shown, in step five, the training set samples F train Input into formula (3) to construct the SVDD model, the model center is recorded as O, the model radius is recorded as R, the constructed SVDD model will include the training samples within the decision boundary as much as possible, and assign a larger slack variable to the samples exceeding the decision boundary, such as Figure 6 As shown, the SVDD method only requires normal samples to establish the processing condition discrimination criterion; where Θ(R,O,ξ) is the SVDD model, C=0.3 represents the penalty factor, ξ=[ξ1,...,ξ K ] T Represents the slack variable; there are a small number of abnormal samples in the training set, and the abnormal samples can be used to optimize the data description constructed by SVDD using formula (4) to improve the anomaly detection accuracy of SVDD; Where K1 represents the number of normal samples, K2 represents the number of abnormal samples, K1+K2=K, C1=K1 / K represents the penalty factor for positive samples, C2=K2 / K represents the penalty factor for negative samples, φ(F i ) represents the kernel function, which provides the SVDD model to be optimized for step 6;

[0029] Figure 4 As shown, step six, based on formula (5), uses the probability density p(F i ) Regularize the SVDD model constructed in step 5 and assign weights to each sample based on probability density. This allocation strategy is not only interpretable, but also can adaptively determine the influence of each sample on the SVDD model according to the distribution of data samples, such as Figure 7 As shown in the figure, it can be seen that the optimized model effectively improves the accuracy of anomaly detection; In addition, for the training set containing a small number of negative samples, the SVDD model is regularized according to formula (6), where p(F j ) represents the estimated probability density of negative samples, and this step provides the optimized SVDD model to be solved for step 7;

[0030]

[0031] Figure 4 As shown, step seven, using the Lagrange multiplier method, as shown in formula (7), the optimized SVDD model established in step six is ​​solved; where α i ≥0,β i ≥0 represents the Lagrange multiplier. According to the Karush-Kuhn-Tucker condition, formula (7) is solved to obtain the expressions of the model center O and the decision radius R, as shown below: The test set constructed in step 3 is input into the optimized SVDD model to identify the abnormality of each test sample. If the distance between the test sample and the model center exceeds the decision radius R, it is marked as abnormal, otherwise it is normal. The accuracy of the detection result is as follows: Figure 8 As shown, it can be found that the optimized SVDD effectively improves the anomaly detection accuracy. This step provides an optimized anomaly detection model for step eight;

[0032] Figure 4As shown, in step eight, the probability density regularized SVDD model obtained in step seven is integrated into the high-speed machining machine tool monitoring platform. With the help of the Internet of Things and the monitoring system platform, the health status of the high-speed machining process can be monitored remotely in real time. By communicating with the data processing center, once an abnormality is detected, the system can promptly issue an alarm or provide corresponding feedback to ensure the safe and stable operation of the high-speed machining machine tool.

[0033] Figure 4 As shown, the present invention is based on the existing SVDD unsupervised anomaly detection method. Under the joint action of related processes, the reliability of high-speed machining can be quickly and effectively evaluated without interrupting the machining process, so as to improve the machining accuracy and efficiency; specifically, the difficulties faced by the actual machining process are fully considered, and only a small number of normal samples are needed to construct the decision boundary of anomaly detection without relying on a large number of balanced samples; the probability of feature samples is estimated in advance, and the noise interference confused in the normal samples can be effectively avoided according to the estimated probability, which improves the accuracy and generalization ability of anomaly detection, so that the constructed anomaly detection method can adapt to actual industrial scenarios; the SVDD model is optimized by the estimated probability, which effectively enhances the description ability of the SVDD model for complex data distribution, improves the detection accuracy on the one hand, and reduces the false alarm rate on the other hand. The optimized SVDD model is integrated into the high-speed machining process monitoring platform to escort the high-speed machining machine tool. The above shows and describes the main features and advantages of the present invention. For those skilled in the art, it is obvious that the present invention is limited to the details of the above exemplary embodiments, and the present invention can be implemented in other specific forms without departing from the spirit or basic features of the present invention. Therefore, the embodiments should be considered in all respects as illustrative and non-restrictive, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is therefore intended that all changes falling within the meaning and range of equivalent elements of the claims are included in the present invention.

[0034] In addition, it should be understood that although the present specification is described according to the implementation mode, the implementation mode does not only include an independent technical solution. This narrative method of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in the embodiments can also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.

Claims

1. A high-speed machining anomaly detection method based on probability optimization support vector data description, characterized in that: The method comprises the following steps: step 1, using vibration sensors installed on the tool spindle and workbench of the machine tool to collect vibration signals in real time during high-speed machining; step 2, extracting features of the collected vibration signals based on AI, which can suppress noise interference on the one hand and improve detection efficiency on the other hand. The extracted features include the root mean square value M in the time domain and the spectrum energy P in the frequency domain, which are recorded as the feature sample set F=[F1,F2,...,F K ] T , where each feature sample is F k ={M k ,P k }; Step 3: Randomly classify the extracted feature samples and establish training sets F train and the test set F test ; Step 4: Use the Gaussian kernel function to train the training set F constructed in step 3 train Probability estimation is performed, and the estimated probability of each sample data is recorded as p(F i ); Step 5, through the training set F train Construct an SVDD model, with the model center denoted as O and the decision boundary denoted as R. The constructed SVDD does not require abnormal samples, and only normal samples are needed to construct a decision boundary for determining whether the high-speed machining process is abnormal. Step six, use the probability density estimated in step four to perform weighted optimization on the SVDD model constructed in step five, and assign weights to each sample based on the probability density. Step seven, use the Lagrange multiplier method to solve the model established in step six, obtain the expressions of the model center O and the decision boundary R, ​​and input the test set constructed in step three into the optimized SVDD model to identify the abnormal conditions of each test sample. Step eight, integrate the probability density regularized SVDD model obtained in step seven into the high-speed machining machine tool monitoring platform of "Friend Machine Technology", so that the monitoring system can communicate and exchange data with the machine tool in real time, and monitor abnormal conditions in the machining process in real time.

2. The high-speed machining anomaly detection method based on probability optimization support vector data description according to claim 1 is characterized in that: In step 1, the vibration signal formula is D = {(x i ,y i ,z i ),(x j ,y j ,z j )},i,j=1,2,...,N.

3. The high-speed machining anomaly detection method based on probability optimization support vector data description according to claim 1 is characterized in that: In step 3, the training set is used to build the SVDD model, and the test set is used to verify the anomaly detection accuracy of the model.

4. The high-speed machining anomaly detection method based on probability optimization support vector data description according to claim 1 is characterized in that: In step 4, the sample probability density at the center of the data feature distribution is large, and the sample probability density at the edge of the data feature distribution is small. The estimated probability density indicates the importance of each sample.

5. The high-speed machining anomaly detection method based on probability optimization support vector data description according to claim 1 is characterized in that: In step 5, if there are a small number of abnormal samples, the abnormal samples can be used to optimize the data description constructed by SVDD to improve the anomaly detection accuracy of SVDD.

6. The high-speed machining anomaly detection method based on probability optimization support vector data description according to claim 1 is characterized in that: In step six, the allocation strategy is not only explainable, but also can adaptively determine the influence of each sample on the SVDD model according to the distribution of data samples. The optimized model effectively improves the detection accuracy.

7. The high-speed machining anomaly detection method based on probability optimization support vector data description according to claim 1 is characterized in that: In step 7, if the distance between the test sample and the model center exceeds the decision radius R, it is marked as abnormal, otherwise it is normal, and the accuracy of the optimization model is statistically calculated.

8. The high-speed machining anomaly detection method based on probability optimization support vector data description according to claim 1 is characterized in that: In step eight, once an abnormality is detected, the system can promptly issue an alarm or provide corresponding feedback to ensure the safe and stable operation of the high-speed machining machine.

Images