A tool monitoring method and system based on data dimensionality increasing adaptive distinguishing machining modes

Abstract

The application discloses a kind of based on data dimensionality upgrading adaptive distinguishing processing mode cutter monitoring method, the method includes the following steps: utilize current sensor to collect the current data on production line main shaft when cutter processing;Detect whether the new current data collected by current sensor is collected into data storage management module in data storage management module;The current data collected is dimensionality upgraded, and the current data is distinguished under different processing modes based on yolov3 model;Current data under different processing modes are classified using deep residual shrinkage network, to determine whether cutter is damaged;According to the wear degree of the whole cutter obtained by judging, the probability of cutter damage is given, and the result is displayed.The application also discloses a cutter monitoring system for implementing the above cutter monitoring method.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of production management method, and relates to a tool monitoring method and system based on data dimensionality increasing adaptive distinguishing machining modes. BACKGROUND

[0002] In the manufacturing industry, a lathe is often indispensable, and a tool is an important component of the lathe. Tool wear is inevitable in the industry. In the machining process, the tool and the machined part interact, and the material particles of the machined part are transferred to the tool, the hardness of the machined part is uneven, and the like, which can cause tool wear. The worn tool can cause various abnormal conditions during machining. These abnormal conditions can cause poor quality and precision of the machined part, which leads to the scrap of the machined part and even the decrease of the precision of the main shaft of the machine tool. These bring certain troubles to the management of product progress and product quality of the factory. Statistics show that the machine tool downtime caused by tool wear, breakage and the like accounts for about 1 / 5-1 / 3 of the total machine tool downtime. However, if a relatively mature tool state detection and wear prediction system is equipped, the downtime can be reduced by about 75%, and the machining efficiency of the machine tool is effectively improved. Therefore, the tool state detection and wear prediction system can effectively reduce the production loss of the workshop, save costs, and improve the machining precision.

[0003] At present, whether the tool is damaged in the factory is determined by workers with rich experience through the sound of the tool during machining. This scheme is too dependent on the experience of the workers, is easily disturbed by subjective factors, and is difficult to apply to large-scale automatic production processes. The intelligent upgrading of the manufacturing industry requires a new, efficient, intelligent and scientific tool state detection and wear prediction system. In recent years, the information technology in China has developed rapidly, and the rapid development and experience accumulation in the fields of artificial intelligence, Internet of Things, sensors and big data make this technology better serve the manufacturing industry.

[0004] Generally, the data collection scheme for tool machining in a laboratory environment is to let the tool continuously cut or drill, and then take down the tool every certain period of time to observe the structure of the tool through a microscope, and determine the wear stage of the tool through professional knowledge. However, this scheme has the following problems:

[0005] 1. Large migration cost.

[0006] 2. Complex actual machining process.

[0007] 3. The tool is discarded in the severe wear stage.

[0008] In the actual operation of the factory, different production lines often bear different tasks, the processing needs may be different, the selected tool models may be different, and the speeds of the machines may also be different; The problem that accompanies is that the subtle differences in tool structure, material, etc. directly cause inconsistent wear curves, and workers cannot make accurate judgments like professional researchers. It is likely that each production line needs professionals to draw wear curves, causing repetitive labor, increasing labor costs, and affecting the processing progress of the factory.

[0009] Currently, the monitoring research of tools mostly stays in the laboratory stage and is difficult to be implemented in the real factory processing environment. There are two types of implementation in actual processing, one is a detection method based on feature extraction + traditional machine learning classifier, and the other is based on deep learning. The former needs people with professional knowledge to extract features, and the universality is poor, and the cost of transferring to a new production line is large; The latter needs to rely on a large amount of data, but in the actual processing process, there are only a small number of negative samples, and the time period required to accumulate data is relatively long. Another point is that the existing two schemes almost do not describe the pre-processing of the collected data in detail. In actual processing and production, the collected data may be lost due to network fluctuations. In actual processing and production, the lathe may have multiple processing modes, and how to distinguish these mode current data is also not well described. SUMMARY

[0010] In order to solve the problems in the prior art, the purpose of the present application is to provide a tool monitoring method based on data dimensionality increasing adaptive distinguishing processing mode, which can be used in industries that need lathe tools for part production and manufacturing; The adaptive in the present application refers to the selection of tool current data in the processing stage through dimensionality increasing according to the whole current data, and the tool state detection is performed through the selected current data in the processing stage.

[0011] In the present application, the sensor technology is applied to obtain the change trend of the current of the machine tool, and the collected data is transmitted to the control platform in real time. The control platform analyzes the obtained data, judges whether the tool is currently worn, predicts whether the tool will be worn in the future, and feeds back to the front-end page in real time, reminding the user to replace the tool or adjust the cutting parameters and other corresponding measures. The above-mentioned scheme not only can improve the processing efficiency of parts, but also can avoid the major damage of the machine tool caused by the fact that the tool breakage is not found by anyone, and through the tool wear prediction, the workers on the production line can be reminded in advance that the tool wear or breakage may occur soon, so that the worn or broken tool can be replaced in time, and the economic benefit of the production line is improved.

[0012] The object of the application can be achieved by the following technical solutions:

[0013] The application provides a tool monitoring method based on data dimensionality increase adaptive distinguishing machining modes, which comprises the following steps:

[0014] Step one, use a current sensor to collect current data on the main shaft during tool machining; specifically:

[0015] According to the specific scene, it is generally divided into two types:

[0016] 1. For a newly built production line, the corresponding interface is reserved in the production line, the main shaft current data can be monitored, and the current data of the main shaft can be directly monitored by an external current sensor.

[0017] 2. For an old production line, the main shaft current data cannot be directly monitored in the production line, and the production line needs to be modified. A current sensor can be installed at the main shaft position (the installed current sensor only needs to detect the current, and there is no special requirement for the installation position on the main shaft), so as to collect current data.

[0018] Step two, detect whether new current data collected by the current sensor is collected into the data storage management module; if new data is collected into the data storage management module, specifically:

[0019] There are two ways:

[0020] 1. If the tool real-time monitoring module communicates with the data collection module, the data collection module sends information to the tool real-time monitoring module after the collection is completed, the tool real-time monitoring module extracts the corresponding current data file, and obtains the current data.

[0021] 2. If the tool real-time monitoring module does not communicate with the data collection module, the tool real-time monitoring module monitors a preset fixed path for storing current data files in the data storage management module in real time, and when it is detected that the fixed path has new current data files, the tool real-time monitoring module extracts the corresponding current data file, and obtains the current data.

[0022] Step three, the collected current data is increased in dimension, and the current data is distinguished under different machining modes based on the yolov3 model; specifically:

[0023] The collected current data is first preprocessed, and the main work includes verifying whether the current data is valid, that is, whether there is effective current data of the tool in the processing process (the current data not processed is invalid). Then the current data is plotted: the collected current data is a time series data (one-dimensional), and it is more convenient and rapid to obtain information from a two-dimensional image, and now the two-dimensional image processing develops rapidly, so the application upgrades the one-dimensional current data of the time series to two-dimensional data to convert it into two-dimensional data, which is convenient for extracting information. According to the time series of the current data as the x-axis and the current data as the y-axis, a two-dimensional image is plotted.

[0024] The yolov3 model used in the application is a target detection model, and in the application, the yolov3 model can judge different processing modes corresponding to different parts of the picture from the two-dimensional data picture of the current data and the time series.

[0025] As shown in Figure 5 , a yolov3 model capable of distinguishing different processing modes in a two-dimensional data picture is pre-trained, and the generated two-dimensional data picture to be detected is put into the yolov3 model for inference, and the model infers the part of the picture corresponding to different processing modes in the picture, and then calculates the current data under different processing modes according to the coordinates in the picture. The yolov3 only gets coordinate information, and then combines the initial current data to get more accurate data. As shown in Figure 6 , the Figure 6 is a schematic diagram of comparing the standard data of a certain processing mode with the actual current data with part missing (the dark color is the standard data, and the light color is the actual data). In the figure, the horizontal coordinate represents the time sequence point, and the vertical coordinate represents the current data. The current data selected by yolov3 under different processing modes is compared by dynamic time warping algorithm (DTW) to judge whether the selection is correct and whether the data is valid. Specifically, the distance between the data selected by yolov3 algorithm and the standard current data is calculated by dynamic time warping algorithm, and the validity of the data and the selected data is judged by whether the distance is within the pre-set threshold range. As shown in Figure 6 , the dynamic time warping algorithm finds similar points in two time series, uses the sum of the distances between all these similar points, called the warping path distance, to measure the similarity between two time series. When the similarity is lower than the pre-set threshold, it is considered to be dissimilar, that is, the selection is incorrect and the data is invalid, otherwise it is similar, that is, the selection is correct and the data is valid.

[0026] As shown in Figure 3As shown, only the second effective data judgment can be performed in step three, the second is to select the data for screening, but the detection efficiency is low, when the first screening is added, some characteristic judgments can improve the detection efficiency.

[0027] Step four, the current data under different processing modes is classified by a deep residual shrinkage network to determine whether the tool is damaged, specifically:

[0028] The current data under different processing modes obtained in step three is extracted for some features: volatility, maximum value, minimum value, mean value, variance, etc., and these features and the current data after sg filtering and noise reduction are put into the residual shrinkage network for classification to determine whether the tool is damaged. The residual shrinkage network structure is as shown in Figure 7 As shown, the input of the network is the current data to be detected, and the output result is whether the tool corresponding to the current data is damaged, RSBU-CS represents a deep residual block sharing threshold between channels, RSBU-CW represents a deep residual block with different thresholds for each channel, BN represents batch normalization, and FC represents a fully connected layer; the training process of the network is as shown in Figure 4 As shown, a round of training is started, a batch of data (including training set and validation set) is taken out from the data set, the model is used to train the training set, the difference between the output of the model and the real data type is calculated, the loss is calculated, and the weight of the model is updated according to the loss. After a round of model training is completed, whether the model converges is observed to determine whether a new round of training needs to be started.

[0029] Step five, result display. According to the wear evaluation of each workpiece, the wear degree of the whole tool is judged, and the probability of tool damage is given.

[0030] In the present application, the judgment of the damage condition of the tool is set according to actual needs. Wear occurs all the time, and the present application is continuous detection. When the wear reaches a certain degree, an alarm is issued. The degree is defined according to the actual needs of the tool, the accuracy of processing, etc., and is also related to other settings of the production line. In actual implementation, the setting conditions are determined according to the actual situation. The present application also proposes a system for realizing the above tool monitoring method, which comprises a data acquisition module, a data storage management module and a tool real-time monitoring module.

[0031] The data acquisition module comprises a machine tool control system, a current sensor and an acquisition module.

[0032] The machine tool control system is used to send a signal to start or end processing. The acquisition module receives the signal sent by the machine tool control system and collects the current data of the tool in the whole processing process through the current sensor.

[0033] The data storage management module is used for storing and managing historical current data and system judgment results.

[0034] The tool real-time monitoring module is used for monitoring the tool wear degree in real time through the data processing method in the application, and displaying the tool state according to the model judgment result.

[0035] Compared with the prior art, the application has the following advantages:

[0036] 1. The migration cost is small, when changing to other production lines or new processing modes, only the marked box in the generated picture is needed, instead of marking thousands of rows of data.

[0037] 2. The tool monitoring method of the application is end-to-end, which can cope with various processing modes, the system first judges the processing mode to which the data belongs, and then uses the model to judge which processing mode the current data belongs to according to the current data.

[0038] 3. Strong anti-interference, in the actual production process, there are inevitably "noise", the addition of sg filter and deep residual shrinkage network makes the method have strong robustness to "noise". BRIEF DESCRIPTION OF DRAWINGS

[0039] Figure 1 It is a processing table and current sensor installation schematic diagram of the embodiment of the application.

[0040] Figure 2 It is a flowchart of the tool monitoring method based on data dimensionality increasing adaptive distinguishing processing mode of the application.

[0041] Figure 3 It is a flowchart of collecting data preprocessing and distinguishing different processing modes by yolov3 of the application.

[0042] Figure 4 It is a training flowchart of the tool monitoring method of the application.

[0043] Figure 5 It is a yolov3 network structure schematic diagram used by the application.

[0044] Figure 6 It is a distance comparison method schematic diagram of the dynamic time warping algorithm used by the application.

[0045] Figure 7 It is a deep residual shrinkage network structure diagram used by the application.

[0046] Figure 8 It is a system component module diagram of the application. DETAILED DESCRIPTION

[0047] The application will be further described in detail in combination with the following specific examples and drawings. The process, conditions, experimental methods, etc. for implementing the application are the general knowledge and common sense in the art, and the application has no special restrictions.

[0048] The application provides a tool monitoring method based on data dimensionality increase adaptive distinction of machining modes. The tool is monitored in real time, and tool damage can be found in time, so that major losses to production are avoided, and many concurrent problems are solved. The application establishes a tool wear monitoring system based on yolov3 and a deep residual shrinkage network, which is used for monitoring the machining state of the tool in real time. The deep residual shrinkage network can comprehensively reflect the tool wear, a threshold alarm mechanism is set, and interaction with manufacturing execution is provided, so that the tool can be replaced in time when the tool wears. The current data is increased in dimension, and the data is adaptively distinguished in different machining modes based on yolov3, so that the migration cost is reduced and the requirement for data volume is reduced

[0049] The design scheme of the tool monitoring method based on data dimensionality increase adaptive distinction of machining modes is as follows:

[0050] Step one, data acquisition. Different ways are needed to collect the current data of the lathe tool in the machining process according to different scenes, which are mainly divided into the following two scenes:

[0051] 1. For a newly built production line, the monitoring spindle current data interface can be reserved, and the current data of the spindle can be directly monitored.

[0052] 2. For the case of production line transformation, the current data of the spindle cannot be monitored in the production line, and a current sensor can be installed at the position of the spindle to collect the current data.

[0053] If both the above two ways can be used, different scenes can be freely selected.

[0054] Step two, whether the collected new current data is collected into the data storage management module. According to different scenes, there are two ways to choose, which are as follows:

[0055] 1. If the tool real-time monitoring module communicates with the data acquisition module, the data acquisition module sends information to the tool real-time monitoring module after the acquisition is completed, the tool real-time monitoring module extracts the current data file, and obtains the current data.

[0056] 2. If the tool real-time monitoring module does not communicate with the data acquisition module, the tool real-time monitoring module stores the current data file in a certain fixed path in the tool real-time monitoring data storage management module, and when it is monitored that there is a new file in the fixed path, the tool real-time monitoring module extracts the current data file to obtain the current data.

[0057] Step three, distinguishing the collected current data under different processing modes. The collected data is first preprocessed, mainly including verifying whether the data is valid, i.e. whether there is current data of the tool in the processing process. If there is current data of the tool, the data is valid. In actual use, it can also be combined with whether it is current data, whether the data size is reasonable, and whether the production line is correct. Then the current data is plotted: the collected current data is a time series data (one-dimensional), and it is more convenient to obtain information from a two-dimensional image. The time series of the current data is taken as the x-axis, and the current data is taken as the y-axis to draw a graph, generating a two-dimensional image. The generated picture is put into the yolov3 model for reasoning, the yolov3 model can reason out the part of the image belonging to different processing modes, and then the coordinates in the image are used to calculate the current data under different processing modes.

[0058] Step four, using a deep residual shrinkage network to classify and determine whether the tool is damaged under different processing modes; specifically:

[0059] The current data under different processing modes obtained in step three is extracted, and some features such as volatility, maximum value, minimum value, mean value, variance, etc. are extracted. These features and current data after sg filtering and noise reduction are put into the residual shrinkage network for classification, and whether the tool is damaged is determined by the output of the residual shrinkage network.

[0060] (5) Result display.

[0061] According to the wear evaluation of each workpiece, the wear degree of the whole tool is judged, and the probability of tool damage is given.

[0062] Embodiment

[0063] In the embodiment of the application, a tool monitoring method based on data dimensionality increasing adaptive processing mode distinguishing is designed according to the principle of intelligent production. The system of the application monitors the processing tool on the assembly line through the linkage of modules such as current sensors, computers, machine tool control systems, etc. Figure 1As shown, a current sensor is added to the spindle position of the machine tool to monitor the current data of the tool during machining, and the data acquisition module monitors the start machining signal and the end machining signal sent by the machine tool control system in real time. The workpiece to be machined is placed on the workbench by the cooperation of the conveying device and the mechanical arm, the machine tool control system sends a start machining signal, the machine tool starts machining, the current sensor monitors the current data of the spindle in real time, and feeds back to the data acquisition module. The data acquisition module monitors the start machining signal sent by the control system in real time, and when the start machining signal sent by the control system is monitored, a new file is created according to certain rules (for example, "production line + model + time point"), and the data obtained from the current sensor is written into the file in real time. When the data acquisition module monitors the end machining signal sent by the control system, stop writing in the file, and send a signal to the tool real-time monitoring module that the spindle current data of the workpiece machining process has been collected. After receiving the information of the data acquisition module, the tool real-time monitoring module obtains the current data file in the public access space.

[0064] After the tool real-time monitoring module obtains the current machining data of the workpiece, it first judges the obtained file, checks whether it is valid data, that is, whether it is current time data, whether the data size is reasonable, whether the production line is correct, etc. If the data is invalid, it will be searched again in the public access space; if it cannot be found after three times, it will be recorded in the log file. After obtaining the valid current data file, the data is plotted with time sequence as the x-axis and current data as the y-axis, and yolov3 algorithm is used to extract the data under different machining modes of the current data. Since yolov3 network extracts candidate boxes from pictures, the corresponding current data needs to be extracted according to the information of the candidate boxes.

[0065] Some features are extracted from the selected data, such as volatility, maximum value, minimum value, mean value, variance, etc. These features and current data after sg filtering and noise reduction (enhanced anti-interference ability) are put into the residual shrinkage network for classification to determine whether the tool wear has reached the degree of affecting product machining (i.e. tool damage). What degree of wear is defined as tool damage needs to be set according to actual needs. Wear occurs all the time and needs to be continuously detected. When the wear reaches "tool damage", an alarm is sent. How to quantify "tool damage" needs to be defined comprehensively according to the actual needs of the tool, the accuracy of machining, etc. The result of the judgment is fed back to the monitoring page, that is, whether the currently monitored tool has "tool damage".

[0066] The protection scope of the present application is not limited to the above embodiments. Changes and advantages that can be thought of by those skilled in the art without departing from the spirit and scope of the present application are included in the present application, and are protected by the appended claims.

Claims

1. A tool monitoring method for distinguishing machining modes based on data dimensionality increasing self-adaptation, characterized in that, The tool monitoring method comprises the following steps: Step one, collect the current data on the main shaft of the production line during tool machining by using a current sensor; Step two, detect whether new current data collected by the current sensor has been collected into the data storage management module; Step three, the collected current data is upgraded, and the current data is distinguished under different machining modes based on the yolov3 model; Step three further comprises: Step 3.1, the collected current data is preprocessed first, and it is verified whether the current data is valid, that is, whether there is current data of the tool in the machining process; Step 3.2, according to the time sequence of the acquisition of the current data as the x axis and the current data as the y axis, draw a two-dimensional image; Step 3.3, the two-dimensional image generated in step 3.2 is put into the yolov3 model for inference, and the yolov3 model infers the part of the graph belonging to different machining modes in the graph, and then calculates the current data under different machining modes according to the coordinates in the graph; The current data selected by yolov3 under different machining modes is compared by dynamic time warping algorithm DTW, to judge whether the selection is correct and the data is valid; that is, the distance between the data selected by yolov3 algorithm and the standard current data is calculated by dynamic time warping algorithm, and the validity of the data and the selected data is judged by whether the distance is within the pre-set threshold range; The dynamic time warping algorithm DTW finds similar points in the actual sequence and the standard sequence of two time sequences, and uses the sum of the distances between all similar points as the warping path distance to measure the similarity between the two time sequences; When the similarity is lower than the pre-set threshold, it is considered to be dissimilar, that is, the selection is incorrect and the data is invalid, otherwise it is similar, that is, the selection is correct and the data is valid; Step four, classify the current data under different machining modes by using deep residual shrinkage network, to judge whether the tool is damaged; In step four, the current data under different machining modes including volatility, maximum value, minimum value, mean value and variance obtained in step three is extracted, and these features and the current data filtered by sg are put into the residual shrinkage network for classification to judge whether the tool is damaged; Step five, according to the obtained wear degree of the whole tool, the probability of tool damage is given, and the result is displayed.

2. The tool monitoring method according to claim 1, wherein In step one, the current data is obtained by directly monitoring through the reserved current monitoring interface on the newly built production line, or by adding current sensors on the old production line.

3. The tool monitoring method of claim 1, wherein, Step two further comprises: if the tool real-time monitoring module communicates with the data acquisition module, the data acquisition module sends information to the tool real-time monitoring module after data acquisition is completed, and the tool real-time monitoring module extracts the corresponding current data file to obtain the current data; If the tool real-time monitoring module does not communicate with the data acquisition module, the tool real-time monitoring module stores the current data file in a preset fixed path in the tool real-time monitoring data storage management module, and when a new current data file is monitored in the fixed path, the tool real-time monitoring module extracts the corresponding current data file to obtain the current data.

4. A tool monitoring system implementing the tool monitoring method according to any one of claims 1 to 3, characterized in that The tool monitoring system comprises a data acquisition module, a data storage management module and a tool real-time monitoring module. The data acquisition module comprises a machine tool control system, a current sensor and an acquisition module; the machine tool control system is used to send a signal for starting or ending machining; the acquisition module receives the signal sent by the machine tool control system and collects current data of the tool in the whole machining process through the current sensor; The data storage management module is used to store and manage historical current data and system judgment results; The tool real-time monitoring module is used to monitor the tool wear degree in real time and display the tool state according to the model judgment result.

Images