A tool abnormality signal recognition method

Abstract

The application provides a tool abnormal signal identification method, which comprises the following steps: firstly, selecting a primary to-be-identified segment and a primary comparison segment from an aligned tool initial signal and a standard template signal; then, selecting a Manhattan distance maximum alignment point as a starting point of the identification segment, which is helpful to redundancy and signal-to-noise ratio improvement; and finally, according to a Manhattan operation result of the identification segment and the comparison segment, judging whether the difference identification segment is abnormal according to a three-sigma principle, so that the robustness of the identification result is ensured.

Description

Technical Field

[0001] This invention relates to the field of industrial control, and in particular to a method for identifying abnormal tool signals. Background Technology

[0002] With the development of sensors, big data, and edge computing devices, predictive maintenance (PHM) and real-time monitoring technologies for machine tools and cutting tools have become possible.

[0003] The common technical architecture currently includes the following components: data acquisition, data preprocessing, feature extraction, model building, and iterative model correction. This architecture typically supports multiple data sources and data types, such as spindle current load signals, vibration signals, sound signals, and image signals.

[0004] The data acquisition process involves collecting data from a monitored object (taking a specific machine tool as an example) via network card / serial port, along with other auxiliary information such as time and equipment operating status. After merging all collected data fields, a data preprocessing step is performed to remove noisy data before sending it to subsequent steps for feature extraction and model building.

[0005] Most commonly used feature extraction and model building methods are based on the form of time-series signals and can be roughly classified into two categories:

[0006] (1) Classical signal processing schemes. For example, the EMD method finds the upper and lower envelopes of a time-series signal through cubic interpolation fitting, then iteratively retrieves the components of the original signal and decomposes it into the sum of the corresponding components. The iteration stops when the convergence boundary is reached, and then the obtained components are used for feature extraction as subsequent analysis indicators. This method requires third-order polynomial interpolation, which is inefficient. When the signal length is slightly long or the signal-to-noise ratio is low, the computational resources consumed are enormous, making it difficult to apply in real-world scenarios.

[0007] (2) Machine Learning Algorithms. Feature extraction and model building for time-series signals can be viewed as a classification problem in machine learning. Existing methods include Support Vector Machines (SVM) and Convolutional Neural Networks (CNN), most of which require perfectly aligned, equal-length time-series data as input. However, in actual production environments, considering factors such as network communication latency and the computing power of acquisition modules and machine tool chips, it is impossible to obtain time-series signals with uniform acquisition intervals. Moreover, training such models requires a large amount of data to ensure reliability. Summary of the Invention

[0008] To address the above problems, this invention proposes a method for identifying abnormal tool signals.

[0009] The main contents of this invention include:

[0010] A tool abnormal signal identification method, comprising:

[0011] Obtaining a tool initial signal transmitted by a machine tool;

[0012] After aligning the tool initial signal with a standard template signal, selecting a primary to-be-identified segment in the tool initial signal and a primary comparison segment corresponding to the primary to-be-identified segment in the standard template signal;

[0013] Comparing the primary to-be-identified segment and the primary comparison segment to obtain an identified segment and a comparison segment;

[0014] According to the Manhattan operation result of the identified segment and the comparison segment, determining whether the identified segment is abnormal according to the three-sigma principle.

[0015] Preferably, the Manhattan operation result of the primary to-be-identified segment and the primary comparison segment comprises:

[0016] Calculating the Manhattan distance of each pair of points of the identified segment and the comparison segment, summing and averaging to obtain the Manhattan distance average;

[0017] Comparing the Manhattan average with a pre-set threshold value to determine whether the difference identified segment is abnormal according to the three-sigma principle.

[0018] Preferably, comparing the primary to-be-identified segment and the primary comparison segment to obtain an identified segment and a comparison segment comprises:

[0019] Calculating the Manhattan distance of each pair of points of the primary to-be-identified segment and the primary comparison segment;

[0020] Selecting the pair of points with the largest Manhattan record as the starting point of the identified segment, and selecting the end point of the primary to-be-identified segment as the end point of the identified segment.

[0021] Preferably, after aligning the tool initial signal with a standard template signal, selecting a primary to-be-identified segment in the tool initial signal and a primary comparison segment corresponding to the primary to-be-identified segment in the standard template signal comprises:

[0022] Windowed sliding is performed on the tool initial signal and the standard template signal, and the distance between the sliding windows is calculated;

[0023] The pair of points with the largest sliding window distance is selected as the starting point of the primary to-be-identified segment, and the size of the sliding window determines the end point of the primary to-be-identified segment.

[0024] Preferably, the generation of the standard template signal comprises the following steps:

[0025] create an input data set, which is the normal running processing data of a product within a set time, the product being composed of N workpieces, the processing data including processing signals of the N workpieces;

[0026] calculate the DTW distance between the N workpieces to obtain a DTW distance matrix of the product;

[0027] respectively calculate the sum of the DTW distances between each workpiece and the remaining workpieces, denoted as a row DTW distance sum;

[0028] take the workpiece dynamic time warping signal corresponding to the minimum value in the N row DTW distance sums as a standard template signal.

[0029] Preferably, the tool initial signal and the standard template signal are aligned through DTW.

[0030] The tool abnormal signal recognition method proposed by the application has the beneficial effects that: the Manhattan distance maximum alignment point of the aligned preliminary to-be-recognized segment and the corresponding preliminary comparison segment of the standard template signal is selected as the starting point of the recognition segment, which helps to eliminate redundancy and improve the signal-to-noise ratio; according to the Manhattan operation result of the recognition segment and the comparison segment, whether the difference recognition segment is abnormal is determined according to the three-sigma principle, thereby ensuring the robustness of the recognition result. BRIEF DESCRIPTION OF DRAWINGS

[0031] Figure 1 The flowchart of the application. DETAILED DESCRIPTION

[0032] The technical solutions protected by the application will be specifically described below with reference to the accompanying drawings.

[0033] Please refer to Figure 1 The application proposes a tool abnormal signal recognition method, which includes the following steps:

[0034] (1) align the obtained tool initial signal and the standard template signal through DTW;

[0035] (2) window and slide the two aligned signals, and calculate the distance between the sliding windows;

[0036] take the position with the maximum sliding window distance as the starting point of the difference interval, and take all the signal data after the starting point as the preliminary to-be-recognized segment, while taking the corresponding signal segment of the standard template signal and the preliminary to-be-recognized segment as the preliminary comparison segment; that is, the bifurcation point of the two signals is selected. Since the monitored object may make multiple processing actions, which is reflected in the signal as multiple similar subsequences, the damage of the monitored object (tool) may occur on any subsequence. Therefore, the starting point selection operation can eliminate the influence of healthy subsequences, improve the signal-to-noise ratio, and thus improve the discrimination accuracy.

[0037] (3) In the present application, after the primary to-be-identified segment is selected, a more accurate identified segment needs to be constructed, specifically, the Manhattan distance of each pair of sites in the primary to-be-identified segment and the primary reference segment is compared, the maximum value of the Manhattan distance is selected as the starting point, and the identified segment and the reference segment are constructed.

[0038] (4) According to the Manhattan operation result of the identified segment and the reference segment, whether the difference identified segment is abnormal is judged according to the three-sigma principle, specifically, the corresponding Manhattan distance of each pair of sites of the identified segment and the reference segment is calculated, and the sum is averaged to obtain the Manhattan distance average; the Manhattan average is compared with the pre-set threshold value, and whether the difference identified segment is abnormal is judged according to the three-sigma principle.

[0039] The Manhattan average is selected to make the calculation result more accurate. If the starting point of the identified segment determined in the previous step is only identified as an abnormal point because there is a single abnormal value in the signal, then the subsequent pairing distance average can spread the influence of the abnormal value to the whole comparison segment, reducing its influence. At the same time, if the real abnormality occurs at the tail of the signal to be judged, the Manhattan distance of the two local signals is magnified because the number of points for averaging is small, and better discrimination effect can be obtained.

[0040] The present application also provides a method for generating a standard template signal, specifically comprising the following steps:

[0041] An input data set is created, the input data set is the processing data of the normal operation of the product within a set time, the product is composed of N workpieces, and the processing data includes the processing signals of the N workpieces;

[0042] The DTW distance between the N workpieces is calculated to obtain a DTW distance matrix of the product, and the size of the DTW distance matrix is N*N; for example, when a product includes three workpieces, the DTW matrix of the product is:

[0043]

[0044] The sum of the DTW distances of each workpiece and the remaining workpieces is calculated respectively, and is recorded as the row DTW distance sum;

[0045] The signal of the workpiece corresponding to the minimum value of the N row DTW distance sums after dynamic time warping is taken as the standard template signal.

[0046] The above only describes the embodiments of the present application, and does not limit the patent scope of the present application, any equivalent structure or equivalent process transformation using the content of the present application specification and the drawings, or direct or indirect application in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A method for identifying abnormal tool signals, characterized in that, include: Acquire the initial tool signal transmitted from the machine tool; After aligning the initial tool signal with the standard template signal, select the primary segment to be identified in the initial tool signal, as well as the primary comparison segment corresponding to the primary segment to be identified in the standard template signal. Compare the Manhattan distances of each pair of points in the initial segment to be identified and the initial comparison segment, select the maximum value of the Manhattan distance as the starting point, and construct the segment to be identified and the comparison segment. Based on the Manhattan operation results of the identification segment and the comparison segment, the three Sigma principle is used to determine whether the identification segment has an anomaly. After aligning the initial tool signal with the standard template signal, the primary segment to be identified in the initial tool signal, and the primary comparison segment corresponding to the primary segment to be identified in the standard template signal are selected, including: Window the tool initial signal and the standard template signal, and calculate the distance between the sliding windows; The alignment point with the largest sliding window distance is taken as the starting point of the primary segment to be identified, and the ending point of the primary segment to be identified is determined by the size of the sliding window. Compare the initial segment to be identified with the initial comparison segment to obtain the segment to be identified and the comparison segment, including: Calculate the Manhattan distance for each pair of sites between the primary segment to be identified and the primary comparison segment; The pair with the largest Manhattan distance is selected as the starting point of the recognition segment, and the end point of the primary segment to be recognized is selected as the end point of the recognition segment.

2. The method for identifying abnormal tool signals according to claim 1, characterized in that, Based on the Manhattan operation results of the primary segment to be identified and the comparison segment, the determination of whether the segment to be identified is abnormal according to the three Sigma principle includes: For each pair of points in the identification segment and the comparison segment, calculate the corresponding Manhattan distance, sum them and take the average to obtain the mean Manhattan distance; The average Manhattan distance is compared with a preset threshold, and the anomaly of the identified segment is determined according to the three Sigma principle.

3. The method for identifying abnormal tool signals according to claim 1, characterized in that, The steps for generating a standard template signal are as follows: Create an input dataset, which is the processing data of the product's normal operation within a set time. The product consists of N workpieces, and the processing data includes the processing signals of the N workpieces. Calculate the DTW distance between N workpieces to obtain the DTW distance matrix of the product; Calculate the sum of the DTW distances between each workpiece and the rest of the workpieces, and record it as the row DTW distance sum; The workpiece dynamic time-normalized signal corresponding to the minimum value among the N rows of DTW distance sums is taken as the standard template signal.

4. The method for identifying abnormal tool signals according to claim 1, characterized in that, The tool initial signal and the standard template signal are aligned via DTW.

Images