Method and device for detecting abnormal torque of main shaft of cutting machine and computer equipment

Abstract

The invention relates to a cutting machine spindle torque anomaly detection method and device and computer equipment. The method comprises the following steps: acquiring technological process data of the cutting machine in a first time range; inputting the technological process data into a pre-trained torque prediction model to obtain torque prediction data in a second time range; torque measured data of the cutting machine in a second time range are obtained; and determining an abnormal detection result of the main shaft torque of the cutting machine based on the torque prediction data and the torque actual measurement data. By adopting the method, the accuracy and the real-time performance of torque anomaly detection can be improved.

Description

Technical Field

[0001] This application relates to the field of detection technology, and in particular to a method, apparatus and computer equipment for detecting abnormal torque of a cutting machine spindle. Background Technology

[0002] Multi-wire cutting is a cutting method that uses the high-speed reciprocating motion of diamond wire to simultaneously cut hard and brittle materials into multiple pieces. Cutting machines using multi-wire cutting typically include two sets of main rollers, each equipped with two spindles: a drive shaft and a driven shaft. The drive shaft is directly driven by a motor, while the driven shaft is connected to the drive shaft via the main rollers and is also driven by the drive shaft. The spindle is the core component of the cutting machine, and its performance is directly proportional to the overall performance of the cutting machine.

[0003] Abnormal spindle load torque is one of the abnormal phenomena in cutting machines. Torque is the force that drives the spindle to rotate. In traditional technology, abnormal spindle load torque is often detected by collecting data through a load torque detection device and then manually conducting irregular inspections. However, the method of judging by observing the torque curve on the industrial control computer interface requires the operator to make judgments based on work experience, which is not objective and lacks a unified judgment standard. The accuracy and real-time performance of abnormal situation judgment are low.

[0004] This shows that the current methods for judging abnormal spindle load torque still suffer from low accuracy and real-time performance. Summary of the Invention

[0005] Therefore, it is necessary to provide a method, device, and computer equipment for detecting abnormal spindle torque of a cutting machine that can improve accuracy and real-time performance in response to the above-mentioned technical problems.

[0006] In a first aspect, this application provides a method for detecting abnormal torque of a cutting machine spindle, the method comprising:

[0007] Obtain the cutting machine's process data within the first time frame;

[0008] The process data is input into a pre-trained torque prediction model to obtain torque prediction data within a second time range;

[0009] Obtain the measured torque data of the cutting machine within the second time range;

[0010] Based on the torque prediction data and the measured torque data, the abnormal detection result of the spindle torque of the cutting machine is determined.

[0011] In one embodiment, acquiring the cutting machine's process data within a first time range includes:

[0012] Retrieve the first process data within a preset time range prior to the current moment stored in the database;

[0013] Obtain the second process data collected at the current moment;

[0014] The process data is obtained based on the first process data and the second process data.

[0015] In one embodiment, before inputting the process data into a pre-trained torque prediction model to obtain torque prediction data within a second time range, the method further includes:

[0016] Obtain historical process data of the cutting machine within a preset historical time range;

[0017] The process data is preprocessed to obtain historical process data;

[0018] Historical process data is used as training samples and input into a machine learning model for training to obtain a torque prediction model.

[0019] In one embodiment, the preprocessing of the historical process data to obtain historical process data includes:

[0020] Time alignment is performed on each type of parameter in the historical process data to obtain time-aligned process data;

[0021] The time-aligned process data is filtered for outliers to obtain steady-state process data;

[0022] Based on preset feature selection rules, correlation analysis is performed on each type of parameter in the steady-state process data to obtain historical process data.

[0023] In one embodiment, the historical process data includes multiple cutting start positions. Data filtering of the time-aligned process data yields steady-state process data, including:

[0024] Based on the cutting start position, the time-aligned process data is batch-grouped to obtain multiple batches of cutting data;

[0025] Based on the multiple batches of cutting data, identify abnormal working conditions and uniform cutting time ranges in the batches of cutting data;

[0026] Based on the abnormal operating conditions and the uniform cutting time range, multiple batches of cutting data are filtered to obtain steady-state process data.

[0027] In one embodiment, before performing correlation analysis on each type of parameter in the steady-state process data based on preset feature selection rules to obtain historical process data, the method further includes:

[0028] The steady-state process data is subjected to absolute value processing and smoothing filtering to obtain the processed steady-state process data.

[0029] In one embodiment, the machine learning model is an extreme gradient boosting model, and the step of using historical process data as training samples and inputting them into the machine learning model for training to obtain the torque prediction model further includes:

[0030] The particle swarm of the extreme gradient boosting model is configured based on a multi-objective particle optimization algorithm;

[0031] Based on the historical process data and the extreme gradient model, the prediction results are obtained;

[0032] Based on the prediction results, the historical process data, and the preset loss function, the particle swarm optimization of the extreme gradient boosting model is adjusted.

[0033] The optimal position of the particle swarm is calculated based on a multi-objective particle swarm optimization algorithm.

[0034] The torque prediction model is determined based on the optimal position of the particle swarm obtained through multiple iterations of training.

[0035] In one embodiment, determining the abnormal detection result of the spindle torque of the cutting machine based on the torque prediction data and the measured torque data includes:

[0036] Calculate the residual between the predicted torque data and the measured torque data;

[0037] If the number of data in the predicted torque data and the measured torque data whose residuals are greater than the residual threshold is greater than a preset number threshold, or the proportion of data whose residuals are greater than the residual threshold is greater than a preset proportion threshold, then the abnormal detection result of the spindle torque is abnormal.

[0038] Secondly, this application provides a device for detecting abnormal torque of a cutting machine spindle, the device comprising:

[0039] The acquisition module is used to acquire the process data of the cutting machine within the first time range;

[0040] The prediction module is used to input the process data into a pre-trained torque prediction model to obtain torque prediction data within a second time range.

[0041] The measurement module is used to acquire the measured torque data of the cutting machine within a second time range;

[0042] The detection module is used to determine the abnormal detection result of the spindle torque of the cutting machine based on the torque prediction data and the measured torque data.

[0043] Thirdly, this application provides a computer device including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method described above.

[0044] The aforementioned method, apparatus, and computer equipment for detecting abnormal spindle torque of a cutting machine acquire process data of the cutting machine within a first time range; input the process data into a pre-trained torque prediction model to obtain torque prediction data within a second time range; acquire measured torque data of the cutting machine within the second time range; and determine the abnormal detection result of the spindle torque of the cutting machine based on the torque prediction data and the measured torque data. This is achieved by using a prediction model to predict torque based on historical process data and comparing it with the measured torque data to determine whether there has been an abnormal deviation in the load torque, thereby realizing the detection of torque anomalies and improving the accuracy and real-time performance of torque anomaly detection. Attached Figure Description

[0045] Figure 1 This is a diagram illustrating the application environment of a cutting machine spindle torque anomaly detection method in one embodiment.

[0046] Figure 2 This is a flowchart illustrating a method for detecting abnormal spindle torque in a cutting machine in one embodiment;

[0047] Figure 3 This is a flowchart illustrating a method for detecting abnormal spindle torque in a cutting machine in another embodiment;

[0048] Figure 4 This is a schematic diagram of the process data preprocessing in one embodiment;

[0049] Figure 5 This is a schematic diagram of the curves for data preprocessing in one embodiment of the cutting process;

[0050] Figure 6 This is a schematic diagram of the torque prediction model training process in one embodiment;

[0051] Figure 7 This is a flowchart illustrating a method for detecting abnormal spindle torque in a cutting machine in another embodiment;

[0052] Figure 8 This is a structural block diagram of a cutting machine spindle torque abnormality detection device in one embodiment;

[0053] Figure 9 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0055] The abnormal torque detection method for cutting machine spindles provided in this application embodiment can be applied to, for example... Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104 or located in the cloud or on other network servers. Terminal 102, through communication with server 104, obtains the cutting machine's process data within a first time range; inputs this process data into a pre-trained torque prediction model to obtain torque prediction data within a second time range; obtains the actual torque measurement data of the cutting machine within the second time range; and, based on the torque prediction data and the actual torque measurement data, determines the abnormal detection result of the cutting machine's spindle torque. Terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. Server 104 can be implemented using a standalone server or a server cluster consisting of multiple servers.

[0056] In one embodiment, such as Figure 2 As shown, a method for detecting abnormal spindle torque in a cutting machine is provided, which is then applied to... Figure 1 Taking terminal 102 as an example, the explanation includes the following steps:

[0057] Step S100: Obtain the process data of the cutting machine within the first time range.

[0058] The process data includes process parameters collected by the cutting machine during operation. For example, the process data may include equipment operating status data, which records the status parameters of each component of the cutting machine during operation, such as bearing housing temperature, cutting fluid flow rate, cooling water pressure, spindle current, voltage, spindle torque, vibration amplitude, etc. The process data also includes equipment operating condition data, which records the position and speed of each component during the process of driving the cutting machine, such as wire mesh speed, spindle speed, material feed speed, material platform position, motor power, etc., which characterize the operating condition of the cutting machine.

[0059] The first time range can be a time range determined based on the current moment. For example, it can be a time range determined by a preset time length based on the current moment. Alternatively, the first time range can be a time range determined by the cutting machine based on multiple marker nodes in historical timelines, with the most recent marker node as the starting point and the current moment as the ending point. For example, the cutting process of the cutting machine can be considered as a series of cutting operations, each including the complete process from tool entry, main cutting to tool exit. Tool entry is the starting stage of the cutting operation. Multiple tool entry moments can be determined as marker nodes, with the most recent tool entry moment as the starting point and the current moment as the ending point to determine the first time range. The first time range can also be determined using other methods for determining the time range of detecting spindle torque; this embodiment does not limit this method.

[0060] Step S200: Input the process data into the pre-trained torque prediction model to obtain torque prediction data within the second time range.

[0061] The torque prediction model can be pre-trained based on historical process data, which can be pre-screened normal process data. This allows the torque prediction model to output torque prediction data under ideal conditions of normal torque load based on the input process data.

[0062] The second time range is a time node or time range after the first time range. For example, a time node can be a preset moment after the first time range; the time range can be a time range determined according to a preset time length, starting from the first moment after the first time range.

[0063] The first and second time ranges can be non-overlapping. In this embodiment, the torque prediction model predicts the torque prediction data for the second time range based on the process data in the first time range.

[0064] Step S300: Obtain the measured torque data of the cutting machine within the second time range.

[0065] The acquisition of process data and the acquisition of measured torque data can be performed synchronously or asynchronously. For example, both process data and measured torque data can be historical data, allowing for the synchronous or asynchronous acquisition of process data within a first time range and measured torque data within a second time range to verify whether there are any anomalies in the spindle load torque at historical moments. In the asynchronous execution scenario, the measured torque data can also be acquired in real time, allowing for the determination of whether there are any anomalies in the spindle load torque at the corresponding moment based on the torque prediction data obtained from the process data within the first time range.

[0066] In this embodiment, the recording formats of the torque prediction data and the measured torque data can be the same or similar. For example, they can use a unified time unit, the same fields, and the corresponding torque unit, so that subsequent processing can be performed based on the torque prediction data and the measured torque data to determine the anomaly detection results.

[0067] Step S400: Based on the torque prediction data and the actual torque measurement data, determine the abnormal detection result of the spindle torque of the cutting machine.

[0068] This can involve comparing predicted torque data and measured torque data to determine the deviation between the two. The deviation can be calculated using methods such as absolute deviation, mean square error, root mean square error, mean absolute error, or relative error; this embodiment does not limit the method. When the deviation exceeds a preset threshold, it can be determined that there may be an abnormality in the spindle load torque within the second time range, and the corresponding abnormality detection result will be output.

[0069] When there is an abnormality in the spindle load torque, different early warning strategies can be triggered according to the degree of abnormality to notify maintenance personnel to check or take corresponding measures to avoid potential equipment damage or production accidents.

[0070] If the abnormal detection result of the spindle load torque is abnormal, and the maintenance personnel determine that the condition is good after inspection, the torque prediction model can be fine-tuned or the deviation calculation parameters can be adjusted based on the process data and torque measurement data used in the abnormal detection result, thereby continuously optimizing the torque prediction model or the accuracy of abnormal detection.

[0071] This embodiment provides a method for detecting abnormal spindle torque of a cutting machine. The method involves acquiring process data of the cutting machine within a first time range; inputting this process data into a pre-trained torque prediction model to obtain torque prediction data within a second time range; acquiring measured torque data of the cutting machine within the second time range; and determining the abnormal detection result of the cutting machine's spindle torque based on the torque prediction data and the measured torque data. The method obtains the torque prediction data by using a prediction model based on historical process data and compares it with the measured torque data to determine whether there is an abnormal deviation in the load torque, thereby achieving torque anomaly detection and improving the accuracy and real-time performance of torque anomaly detection.

[0072] In one embodiment, acquiring the cutting machine's process data within a first time range includes:

[0073] Retrieve the first process data within a preset time range prior to the current moment stored in the database;

[0074] Obtain the second process data collected at the current moment;

[0075] Based on the data from the first process and the data from the second process, the process data is obtained.

[0076] In this embodiment, the process data and the measured torque data are acquired asynchronously.

[0077] Historical process data within a preset time range prior to the current moment will be stored in a database, referred to in this embodiment as first process data. For example, it could be stored in a database such as... Figure 1 In the data storage system shown, terminal 102 communicates with server 104 via a network to obtain the first process data from the data storage system. In another embodiment, the first process data may also be stored in an industrial control computer and can be obtained by accessing the data of the industrial control computer.

[0078] The second process data can be the process data collected in real time at the current moment, which is referred to as the second process data in this embodiment.

[0079] Based on the first process data and the second process data, the process data can be obtained by splicing the first process data and the second process data.

[0080] Furthermore, the cutting machine spindle torque abnormality detection method of this application can be repeatedly executed. Before entering the next moment, the second process data collected at the current moment can be stored in the database and recorded as part of the first process data. After entering the next moment, the process data of the next moment is collected in real time and used as the second process data.

[0081] This embodiment provides a method for detecting abnormal spindle torque in a cutting machine. By acquiring process data stored in a database and real-time collected process data, process data within a first time range can be obtained. This ensures the continuity, integrity, and timeliness of the process data, thereby improving the efficiency and accuracy of the torque prediction data output by the torque prediction model.

[0082] In one embodiment, before inputting the process data into a pre-trained torque prediction model to obtain torque prediction data within a second time range, the method further includes:

[0083] Obtain historical process data of the cutting machine within a preset historical time range;

[0084] The process data is preprocessed to obtain historical process data;

[0085] Historical process data is used as training samples and input into a machine learning model for training to obtain a torque prediction model.

[0086] Among them, obtaining historical process data of the cutting machine within a preset historical time range can be process data collected from a historical database over a longer period of time as historical process data.

[0087] Preprocessing the process data, for example, can involve data cleaning, feature extraction, and data normalization, so that historical process data can be used as training samples to be input into a machine learning model for training.

[0088] The machine learning model can be an existing machine learning algorithm, such as linear regression, support vector machine, random forest, long short-term memory network, etc. The machine learning algorithm used can also be determined according to the actual needs of the operator, which is not limited in this embodiment.

[0089] Historical process data is used as training samples and input into the machine learning model. This can be achieved by dividing the preprocessed historical process data into training, validation, and test sets, and using cross-validation to evaluate model performance, prevent overfitting, and adjust model parameters to optimize prediction results. After training, the model's predictive performance is evaluated using a test set, employing metrics such as mean squared error, root mean square error, and mean absolute error to ensure the model has sufficient generalization ability to accurately predict torque changes.

[0090] This embodiment provides a method for detecting abnormal spindle torque in a cutting machine. By preprocessing historical process data within a historical time range and inputting it as training samples into a machine learning model, a more accurate torque prediction model can be obtained, thereby improving the accuracy of abnormal spindle torque detection.

[0091] In one embodiment, the preprocessing of the historical process data to obtain historical process data includes:

[0092] Time alignment is performed on each type of parameter in the historical process data to obtain time-aligned process data;

[0093] By filtering out abnormal data from the time-aligned process data, steady-state process data is obtained.

[0094] Based on preset feature selection rules, correlation analysis is performed on each type of parameter in the steady-state process data to obtain historical process data.

[0095] Since different sensors or data sources may have inconsistent timestamps (e.g., different sensors have different sampling frequencies or data recording times), time alignment of each parameter in the historical process data can be performed to obtain time-aligned process data, ensuring that all data is aligned to the same time reference. Time alignment can employ one or more processing methods such as data resampling, timestamp matching, and interpolation. In one specific embodiment, the parameters with lower sampling frequencies can be sampled and interpolated based on the highest sampling frequency in the historical process data, and the features of other parameters can be aligned sequentially. For missing time points, the average of the two most recent data points is calculated to fill the gaps.

[0096] To filter out abnormal data in the time alignment process data, some time alignment process data whose parameters obviously do not conform to the normal process can be screened out. In addition, if there are abnormalities such as broken wires, skipped wires, excessive bow, vibration jamming, etc., the time alignment process data corresponding to the above abnormalities should also be screened out, so as to avoid the torque prediction model trained to predict non-ideal torque prediction data.

[0097] Based on preset feature selection rules, correlation analysis is performed on each type of parameter in the steady-state process data to obtain historical process data. This can involve calculating the correlation between each type of parameter in the steady-state process data and torque, and using parameters with a correlation higher than a certain level as training samples. This can effectively optimize model performance and reduce dimensionality. For example, correlation coefficient analysis methods such as Pearson correlation coefficient and Spearman correlation coefficient can be used to calculate the correlation. Alternatively, methods such as mutual information and recursive feature elimination can be used to evaluate the correlation strength between each parameter and torque. This paper does not limit the scope of these methods.

[0098] This embodiment provides a method for detecting abnormal torque on a cutting machine spindle. By aligning time and filtering abnormal data, and retaining highly correlated parameters as training samples for a machine learning model, the training efficiency and prediction accuracy of the torque prediction model can be effectively improved.

[0099] In one embodiment, the historical process data includes multiple cutting start positions. Data filtering of this time-aligned process data yields steady-state process data, including:

[0100] Based on the cutting start position, the time-aligned process data is batch-grouped to obtain multiple batches of cutting data;

[0101] Based on the multiple batches of cutting data, identify abnormal working conditions and uniform cutting time ranges in the batches of cutting data;

[0102] Based on the abnormal working condition and the uniform cutting time range, the cutting data of multiple batches were filtered to obtain steady-state process data.

[0103] The cutting start position can be either the location within the time window at the start of the cutting process or the cutting start moment. It's understood that the cutting process of a cutting machine involves multiple cutting operations. The cutting start moment can be the entry point of each cutting operation. Based on the cutting start position, the time-aligned process data is batch-grouped. This allows the time-aligned process data to be divided into batches according to each cutting operation, enabling independent analysis of each cutting task and further improving the accuracy of torque prediction.

[0104] Identifying abnormal operating conditions in batch cutting data and filtering the batch cutting data based on these abnormal conditions can involve identifying whether there are abnormal operating conditions such as equipment failure, material jamming, unplanned downtime, torque surge, and current fluctuation in the batch cutting data. This allows batch cutting data containing these abnormal conditions to be filtered out, avoiding the use of batch cutting data containing abnormal conditions as training samples, thereby improving the accuracy of the torque prediction model's output results.

[0105] During the cutting process, the cutting machine usually adopts frequent left and right reversal to improve the cutting ability. However, the cutting speed will vary with time during each cutting operation due to frequent reversal. However, the period that plays the main cutting role is the uniform cutting stage several seconds after the tool is put into operation. Therefore, it is necessary to identify the uniform cutting time range in the batch cutting data and filter the batch cutting data based on this uniform cutting time range. This can be done by retaining only the batch cutting data within the uniform cutting time range to obtain steady-state process data.

[0106] This embodiment provides a method for detecting abnormal spindle torque of a cutting machine. By dividing the time-aligned process data into batches based on the cutting start position, and then filtering the batch cutting data according to the abnormal working conditions and the uniform cutting time range, abnormal and atypical working conditions that may distort the model learning can be removed, thereby improving the accuracy of the torque prediction model prediction data.

[0107] In one embodiment, before performing correlation analysis on each type of parameter in the steady-state process data based on preset feature selection rules to obtain historical process data, the method further includes:

[0108] The steady-state process data is processed by absolute value processing and smoothing filtering to obtain the processed steady-state process data.

[0109] Absolute value processing can be used to apply absolute values ​​to directional parameters in the steady-state process data. These directional parameters are those strongly correlated with direction in the steady-state process data, such as the direction of motor current and the direction of wire mesh movement. In process data, the positive and negative signs are often used to reflect the direction. However, when building a torque prediction model, the positive and negative signs can cause a step change in the data. Absolute value processing simplifies the model learning task, eliminates directional interference, and allows the focus to be placed on the impact of parameter magnitude changes on the results.

[0110] Smoothing filtering of the processed data can avoid data fluctuations caused by sampling errors of parameters such as torque or by the decline in the lifespan of the equipment itself, thereby improving the signal-to-noise ratio. In a specific embodiment, the Savitzky-Golay smoothing filtering method can be used, which performs polynomial fitting on the collected net weight curve through a sliding window of length window_length to eliminate random errors caused by sampling accuracy.

[0111] This embodiment provides a method for detecting abnormal torque on a cutting machine spindle. By processing the absolute values ​​of directional parameters and smoothing the steady-state process data, the stability of the data can be enhanced, thereby improving the accuracy and robustness of the torque prediction model.

[0112] In one embodiment, the machine learning model is an extreme gradient boosting model. The process of using historical process data as training samples and inputting it into the machine learning model to obtain the torque prediction model further includes:

[0113] The particle swarm for this extreme gradient boosting model is configured based on a multi-objective particle optimization algorithm;

[0114] Based on the historical process data and the extreme gradient model, the prediction results are obtained;

[0115] Based on the prediction results, the historical process data, and the preset loss function, the particle swarm optimization of the extreme gradient boosting model is adjusted.

[0116] The optimal position of the particle swarm is calculated based on a multi-objective particle swarm optimization algorithm.

[0117] The torque prediction model is determined based on the optimal position of the particle swarm obtained through multiple iterations of training.

[0118] Extreme Gradient Boosting (XGBoost) is an efficient gradient boosting decision tree algorithm that iteratively adds weak learners, such as CART trees (Classification and Regression Trees), to gradually optimize the prediction model.

[0119] In this embodiment, a multi-objective particle swarm optimization (MOPSO) algorithm is used to configure the particle swarm of the extreme gradient boosting model, and the hyperparameters of the model are adjusted using MOPSO. The particle swarm can be each solution in the MOPSO algorithm, i.e., a set of hyperparameter configurations, with each particle searching for the optimal solution in the search space. The initial particle swarm can be set by the operator or randomly generated. Then, based on historical process data, the extreme gradient boosting model with the current particle swarm configuration generates prediction results.

[0120] Based on the historical process data and the extreme gradient model, the prediction result is obtained. This can be achieved by using the historical process data as training samples and inputting iterative training into the extreme gradient model. In each training round, the extreme gradient model learns the characteristics of the current particle swarm optimization data and gradually builds a weak classifier to improve prediction accuracy. Based on the current configuration of the extreme gradient model and the historical process data, the extreme gradient model generates a torque prediction result and compares it with the actual torque value. The historical process data includes the spindle torque at multiple time points; the prediction result can be a spindle torque prediction for the next time point or within the next time range, based on historical process data over a partial time range.

[0121] Based on the prediction results, the historical process data, and the preset loss function, the particle swarm of the extreme gradient boosting model is adjusted. This can be done by evaluating the performance of the current extreme gradient boosting model using the preset loss function based on the difference between the prediction results and the actual values, and then feeding the evaluation results back to the multi-objective particle optimization algorithm to adjust the position of each particle in the particle swarm.

[0122] After multiple iterations, the multi-objective particle swarm optimization algorithm calculates the optimal position of the particle swarm based on the information collected in all iterative training. This optimal position represents the set of optimal or near-optimal hyperparameters found in the search space. Thus, the torque prediction model can be determined based on the optimal position of the particle swarm obtained from multiple iterative training.

[0123] This embodiment provides a method for detecting abnormal torque on a cutting machine spindle. By employing a multi-objective particle swarm optimization algorithm to guide the construction of an extreme gradient boosting model, it is possible to train a torque prediction model with higher accuracy and generalization ability in complex process data, thereby improving the accuracy and efficiency of torque prediction data.

[0124] In one embodiment, the determination of the abnormal detection result of the spindle torque of the cutting machine based on the torque prediction data and the measured torque data includes:

[0125] Calculate the residual between the predicted torque data and the measured torque data;

[0126] If the number of data in the predicted torque data and the measured torque data whose residuals are greater than the residual threshold is greater than the preset number threshold, or the proportion of data whose residuals are greater than the residual threshold is greater than the preset proportion threshold, then the abnormal detection result of the spindle torque is abnormal.

[0127] The calculation of the residual between the torque prediction data and the measured torque data can be performed by calculating the residual between the torque prediction data and the measured torque data at the same time point or data sampling point. If the second time range includes multiple times, the residual of the data at multiple time points or data sampling points can be calculated for evaluation.

[0128] When the number of time points or data points with residuals greater than the residual threshold exceeds the preset quantity threshold, or when the proportion of time points or data points with residuals greater than the residual threshold to the total number of time points or data points exceeds the preset proportion threshold, the abnormal detection result of the spindle torque is abnormal.

[0129] This embodiment provides a method for detecting abnormal spindle torque in a cutting machine. By comparing predicted torque data and measured torque data through quantitative analysis, it can determine whether the load torque is deviating, thereby determining the abnormal detection result of the spindle torque and improving the accuracy and timeliness of abnormal detection.

[0130] To more clearly illustrate the technical solution of this application, a detailed embodiment is also provided.

[0131] In one embodiment, a method for detecting abnormal spindle torque of a cutting machine is provided, applied to a multi-wire cutting machine. This method requires the use of a pre-trained torque prediction model, and the training steps of the torque prediction model include:

[0132] Obtain historical process data of the cutting machine within a preset historical time range.

[0133] Among them, such as Figure 3 As shown, in a multi-wire cutting machine, sensors are arranged at corresponding positions on the machine body to collect and store process data in real time, such as... Figure 4 As shown, the process data includes raw process parameters, which are the complete process parameters collected by various sensors during the operation of the multi-wire cutting machine. These include equipment operating status data, such as bearing housing temperature, cutting fluid flow rate, cooling water pressure, spindle current, and torque; and equipment operating condition data, such as wire mesh speed, spindle speed, material feed rate, and material platform position, which characterize the slicing machine's operating condition. The historical data used to establish the torque prediction model requires complete operating condition information, including the complete process of the material platform from blade entry, main cutting, to blade exit. Among these, wire mesh speed, feed rate, and feed relative zero point position all change dynamically with cutting.

[0134] The raw process parameters collected can be stored in a MySQL database, with each multi-wire cutting machine as a unit. Data is extracted and a database model is constructed based on identifying fields. Data format and constraint relationships are set according to the characteristics of parameter fields. For example, machine type and batch are used as constraints for the latest data in the basic data. The changes in the equipment are monitored in real time based on fields such as batch number and status. Relevant parameters for anomaly detection are extracted from the dynamically changing data and stored in real time on the associated table.

[0135] The process data is preprocessed to obtain historical process data.

[0136] The raw process data acquired by multi-wire EDM machines is characterized by high dimensionality, dynamic changes, diverse types, and uneven sample distribution, making it unsuitable for direct training of machine learning models. Therefore, preprocessing is necessary to transform the data into a more usable training sample set. This requires alignment, grouping, smoothing, filtering, and feature extraction of the raw data to construct training samples. Specifically, the processing includes: parameter time alignment, batch grouping and calibration, normal batch filtering, steady-state condition filtering, symmetric parameter absolute value processing, smoothing filtering, and feature selection.

[0137] For parameter time alignment, such as Figure 4As shown, process parameters are aligned by time. Addressing the issue of sampling frequency differences in the original data, for parameters with lower sampling frequencies, upsampling is performed using the highest standard. This involves employing upsampling interpolation to preserve the characteristics of high-frequency parameters. Furthermore, the parameters are arranged chronologically according to the highest sampling frequency parameter, aligning the times of other parameters sequentially. Missing time points are filled by the average of the two nearest preceding and following data points. For example, for parameters with drastic high-frequency changes such as spindle speed and wire mesh speed, the sampling interval is 1 second; for parameters with slow changes such as cutting fluid temperature and bearing housing temperature, the sampling interval is 5 seconds. Therefore, the first sampling points of the high-frequency and low-frequency parameters are aligned. Then, the second to fifth sampling points of the low-frequency parameters are interpolated and aligned with the second to fifth sampling points of the former, and so on, until all parameters are edited to the same length.

[0138] For batch grouping calibration, which involves grouping the data collected by the multi-wire cutting machine into batches, in this embodiment, batch grouping calibration can be based on the relative zero point position of the feed. Specifically, the zero point of the relative feed position of each batch can be found by sliding time steps. The relative feed position can be a position where the silicon rod is about 1 mm away from the wire mesh. Then, the continuous samples are cut into subsets through each zero point. Each subset must contain the complete entry, main cut and exit processes. Finally, the data points in each subset where the relative feed position is positive and monotonically increasing are counted as a batch group. Each batch group contains the complete dynamic characteristics of each cut and corresponds one-to-one with the label of whether it is abnormal or not.

[0139] For normal batch screening, the history of whether there are broken lines, double-cuts, sudden changes, etc. can be used. Jumping lines, excessive bowing of the wires, vibration jamming, etc. can also be regarded as abnormal situations in the cutting process. Abnormal batches need to be removed and normal batches are selected. That is, groups that do not have long-term downtime during the cutting process and do not have double-cuts after the cutting is completed are considered normal batches.

[0140] For steady-state condition screening, the gradient of wire mesh speed can be used. Since reciprocating multi-wire cutting processes employ frequent left-right reversals to improve cutting capability, the data collected under these non-steady-state conditions exhibits a high degree of periodicity correlated with wire mesh speed. The actual phase contributing to cutting capability is the several-second period of uniform, stable speed after the reversal. Therefore, steady-state condition screening is necessary. Specifically, this can be achieved by calculating the gradient of wire mesh speed and selecting multiple data points where the speed gradient is 0 but the speed value is not 0, as steady-state condition samples. Figure 5 As shown in the figure, curve ① is the original torque curve for a single group. After screening under steady-state conditions, the torque curve of the uniform and stable stage is retained, and the steady-state torque curve as shown in curve ② can be obtained.

[0141] For handling the absolute values ​​of symmetry parameters, one approach is to convert symmetry parameters related to commutation into absolute values. Process data contains parameters related to the wire mesh cutting direction, such as motor torque and speed. During data acquisition, these are often stored as positive and negative values ​​respectively, based on whether the spindle rotates clockwise or counterclockwise, using positive and negative signs to represent direction. Therefore, it is necessary to process direction-related parameters into absolute values ​​to avoid abrupt changes in data before and after commutation, which could affect modeling and analysis. For example... Figure 5 As shown, curve ③ is the steady-state torque curve after absolute value processing.

[0142] For smoothing filtering, due to sampling errors in torque and the degradation of equipment lifespan, the collected data may fluctuate. Therefore, smoothing processing is required through signal filtering to improve the signal-to-noise ratio. Specifically, this involves smoothing the torque sampling curve. The Savitzky-Golay smoothing filter can be used, which performs polynomial fitting on the collected net weight curve through a sliding window of length `window_length` to eliminate random errors caused by sampling accuracy. Due to the boundary effect of the Savitzky-Golay filter, sample points at both ends of the filtered curve, within (window_length-1) / 2, need to be discarded. The load torque curves at each stage obtained through the above data processing flow are shown in Figure X. It can be seen that the originally periodically oscillating and frequently fluctuating curve is finally transformed into a relatively smooth curve with a clear trend, improving the efficiency of the model in extracting data features. Figure 5 As shown, curve ④ is the steady-state torque curve after smoothing and filtering.

[0143] Feature selection can be based on physical meaning and correlation analysis. Since multi-wire EDM process data is hundreds of dimensions, high-dimensional process data will increase the training burden on the model. Therefore, feature selection methods can be used to filter out parameters more relevant to the task and eliminate useless parameters, improving modeling efficiency and accuracy. In this embodiment, the input parameters of the torque prediction model can be determined by combining experience and correlation analysis. Experience can be used to identify parameters that affect torque from a physical mechanism perspective, such as feed rate, material-wire contact area, and wire speed. Correlation analysis can calculate the correlation coefficient between each parameter and torque from a data perspective. Pearson correlation coefficient, Spearman correlation coefficient, etc., can be used for correlation analysis, and several parameters with higher correlation coefficients can be selected as the independent variables input to the model.

[0144] By preprocessing the data and determining the input features, we can obtain training samples, including features and labels, i.e., historical process data, for training machine learning models.

[0145] like Figure 6 As shown, historical process data is used as training samples, which include features and labels. The samples are divided into several groups proportionally, with one group serving as the test set and the rest as the training set. These are then input into a machine learning model for training to obtain a torque prediction model. Specifically, this can be constructed using an XGBoost torque prediction model based on multi-objective particle swarm optimization, employing various methods to improve prediction performance and generalization ability.

[0146] In this embodiment, the model parameters can be optimized based on the MOPSO algorithm, with the training set loss function and the difference between the training set and test set loss functions as the objectives. The machine learning model can be based on an extreme gradient boosting model, combined with a multi-objective particle swarm optimization algorithm to determine the optimal model hyperparameters, and regularization and other methods can be used to avoid overfitting, thereby improving the model's generalization ability. It is understood that determining the optimal model hyperparameters is not limited to the multi-objective particle swarm optimization algorithm; it can also be achieved using hyperparameter optimization algorithms such as ant colony optimization.

[0147] For the construction of the extreme gradient boosting model, the key influencing factors selected in the feature selection stage can be used as input variables, i.e., feature = {f1, f2, f3, ..., fn}, which include, but are not limited to, cutting wire mesh speed, material platform feed relative zero point position, feed speed, material properties such as bar length and contact area, cutting fluid properties such as temperature and flow rate, coolant properties such as temperature, flow rate and pressure, main roller service life such as working hours, etc., where n is the number of features, the torque filtering curve is used as the observed dependent variable, and the relationship between the input independent variables and the output dependent variables is fitted by a continuously split binary tree to generate a nonlinear regression model.

[0148] For the choice of loss function, the SquareLoss function can be used, which aims to minimize the squared error between the predicted torque and the actual torque under normal uniform cutting conditions. For example, the training set samples are ((X... i ), y i ), where (X i ) is the eigenvector, y i The objective function of the torque prediction model is: (labeled as ) in, Ω(f) represents the predicted value, and Ω(f) represents the regularization term.

[0149] In this embodiment, the parameters of the extreme gradient boosting model can be configured based on particle swarm optimization. The particle swarm can be determined by a multi-objective particle swarm optimization algorithm, whose optimization direction is to reduce the loss function value and improve the generalization performance of the model. Specifically, this includes: (1) defining the multi-objective function to be optimized and the loss function, which can be one or more of the following: root mean square error, mean absolute error, regression coefficient, etc.; (2) initializing the particle swarm, where each particle represents a combination of parameters of an extreme gradient boosting model, including but not limited to the learning rate (Learning_Rate), maximum depth (Max_Depth), and the number of trees (base learners) (n_estimators). Each particle has a current velocity and position vector, which can be obtained by random allocation; (3) updating particle position and velocity: training the extreme gradient boosting model based on the current position and velocity of the particles, calculating the values ​​of the two objective functions, and updating the individual best position and global best position of the particles; (4) selecting the global optimal solution: after each iteration, selecting the set of Pareto optimal solutions from the best positions of all particles. When the maximum number of iterations is reached or the stopping condition is met, the optimization algorithm is terminated, and the globally optimal hyperparameter combination is output, thereby determining the optimal hyperparameter set and training the XGBoost prediction model.

[0150] In this embodiment, Ridge regularization, limiting leaf node weights, subsampling, column sampling, and early stopping are used to control model complexity and avoid overfitting to training set features during model learning.

[0151] To use Ridge regularization, the sum of squares of the weight coefficients of all binary tree leaf nodes can be added to the monitoring loss function during model training, as shown in the formula: Where Ω(f) represents the model complexity, i.e., the regularization term, and T represents the number of leaf nodes in the tree. λ is the weight of the j-th leaf node, λ is the regularization coefficient used to control the influence of the regularization term, and γ is used to control the influence of the number of leaf nodes.

[0152] Model complexity can be controlled and the risk of overfitting reduced by limiting the maximum and minimum values ​​of leaf node weights, subsampling, column sampling, and early stopping. Subsampling controls the number of training samples per tree, thereby reducing model variance; column sampling controls the number of features per tree, thereby reducing the correlation between features; early stopping controls the convergence speed of the loss function, terminating training prematurely when the convergence speed approaches saturation.

[0153] The model's prediction performance and generalization ability are tested using a test set. Further, cross-validation can be employed, dividing the training samples into several subsets. One subset is selected alternately as the test set, while the others are used as the training set for training and testing. The model parameters are adjusted using the optimization methods described above until the optimal result is achieved. For example, when the regression coefficient R_score for the test set reaches 0.95, the trained torque prediction model is obtained.

[0154] After the torque prediction model is trained, every unit time T s To acquire the cutting machine's process data within the first time frame. Furthermore, such as... Figure 7 The image shows the real-time data collected from the current moment and the period preceding it in the process of a multi-wire cutting machine, i.e., a multi-dimensional time series.

[0155] This process requires filtering features from the multidimensional time series. Parameters in the process data include those with high relevance identified during the previous feature selection phase, including but not limited to cutting wire mesh speed, material platform feed speed, feed relative zero point position, material properties, cutting fluid properties, spindle motor parameters, cooling water properties, bearing housing parameters, wire mesh tension, and tension mechanism position. These parameters are stored sequentially in time units. The process data can be preprocessed by adjusting parameter time alignment, symmetrical parameter absolute value conversion, and smoothing filtering according to requirements. The first time range of process data may include the current time t. r The acquired process data, and the length of time T before the current moment. l Historical process data.

[0156] The process data is input into the torque prediction model, which outputs a prediction sequence. This prediction sequence is the torque prediction data for the second time range.

[0157] Obtain the measured torque data of the cutting machine within the second time range.

[0158] Based on the predicted torque data and the measured torque data, the abnormal detection result of the spindle torque of the cutting machine is determined. This can involve comparing the residual between the predicted torque data and the actual torque data, i.e., calculating the residual between the predicted sequence and the measured sequence. The proportion or number of variation points exceeding a threshold within the detection window is calculated; if it exceeds a certain value, an abnormal alarm message is generated: when the number of samples with residuals exceeding a set control limit within the sampling window exceeds a preset number, or when the proportion of these samples to the total number of samples exceeds a preset proportion, it is determined that the load torque is abnormal. The possible causes of the abnormality are analyzed based on the magnitude of the measured and predicted values ​​combined with experience.

[0159] It is understood that in this embodiment, real-time online anomaly detection and parameter updates can be based on a sliding window and the 3σ criterion.

[0160] Furthermore, the aforementioned time length T l For unit time T s Twice the number of samples, meaning that between adjacent time series, the last 50% of the data in the first series is the same as the first 50% of the samples in the second series, to achieve sample overlap. This prevents the loss of information caused by the separation of samples between two series under non-overlapping sampling, which can lead to the segmentation of abnormal features and reduce the false negative rate.

[0161] Furthermore, dynamic control thresholds are determined based on the 3σ criterion: when the sample residuals are normally distributed, the control limits for the anomaly detection criteria can be set as dynamic 3σ criteria to adapt to state changes caused by batch-to-batch formula changes and equipment wear in multi-wire cutting machines. Specifically, the residuals between predicted and measured values ​​of historical data within a recent preset time period can be selected, and their mean μ and standard deviation σ can be calculated. The upper and lower control limits are then determined using UCL = μ + 3σ and LCL = μ - 3σ. In actual production, the control limits can also be reasonably scaled down based on detection sensitivity. For example, UCL = p × (μ + 3σ), where p is the scaling ratio. When p is greater than 1, it indicates a higher tolerance and lower alarm rate for the anomaly detection algorithm; when p is less than 1, the opposite is true.

[0162] Furthermore, when the anomaly detection algorithm detects an anomaly, it can check whether there are a certain proportion or number of consecutive mutation points within the detection window before issuing an alarm. This differs from the immediate alarm method and can avoid false alarms caused by a single mutation point due to random errors. In addition, it also includes judging the time difference between the alarm information and previous similar anomaly warning records, and judging whether the time difference exceeds a certain duration. Similar anomalies will not be warned repeatedly in the short term, and it can also avoid the situation where the same anomaly is repeatedly alarmed without being processed.

[0163] Furthermore, this method also includes anomaly warning, push notification, and storage. Warning information can be manually verified, and model parameters updated based on whether a false alarm has occurred. When a warning is generated, it can be stored in the anomaly record table of the database and promptly pushed to the on-site machine control system to inform equipment management personnel to investigate. If the investigation results differ from the detection results, the detection record is modified, and the data is marked as normal for updating model parameters; if the investigation result is abnormal, a reasonable maintenance strategy is formulated to ensure the stable operation of the multi-wire cutting machine.

[0164] This embodiment provides a method for detecting abnormal torque on a cutting machine spindle. The method uses a prediction model to obtain predicted torque data based on historical process data, and compares this predicted torque with actual measured torque data to determine if there is an abnormal deviation in the load torque. This achieves torque anomaly detection, improving the accuracy and real-time performance of torque anomaly detection. By acquiring process data stored in a database and real-time collected process data, the process data for the first time range is obtained, ensuring the continuity, integrity, and timeliness of the process data, thereby improving the efficiency and accuracy of the torque prediction model's output torque prediction data. By preprocessing historical process data within the historical time range and using it as training samples in a machine learning model, a more accurate torque prediction model can be obtained, improving the accuracy of spindle torque anomaly detection. Through time alignment and anomaly data filtering, retaining highly correlated parameters as training samples for the machine learning model effectively improves the training efficiency and prediction accuracy of the torque prediction model. By batching time-aligned process data according to the cutting start position, and then filtering the batch cutting data based on abnormal working conditions and uniform cutting time range, abnormal and atypical working conditions that may distort model learning can be removed, thereby improving the accuracy of torque prediction model prediction data. By processing the absolute values ​​of directional parameters and smoothing the steady-state process data itself, data stability can be enhanced, thus improving the accuracy and robustness of the torque prediction model. By employing a multi-objective particle swarm optimization algorithm to guide the construction of the extreme gradient boosting model, a torque prediction model with stronger accuracy and generalization ability can be trained on complex process data, thereby improving the accuracy and efficiency of torque prediction data. By quantitatively comparing torque prediction data and measured torque data, it is possible to determine whether the load torque is deviating, thus confirming the anomaly detection results of the spindle torque, improving the accuracy and timeliness of anomaly detection.

[0165] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps.

[0166] Based on the same inventive concept, this application also provides a cutting machine spindle torque anomaly detection device for implementing the above-described cutting machine spindle torque anomaly detection method. The solution provided by this device is similar to the implementation described in the above method; therefore, the specific limitations in one or more embodiments of the cutting machine spindle torque anomaly detection device provided below can be found in the limitations of the cutting machine spindle torque anomaly detection method described above, and will not be repeated here.

[0167] In one embodiment, such as Figure 8 As shown, a cutting machine spindle torque abnormality detection device is provided, comprising: an acquisition module 100, a prediction module 200, a measurement module 300, and a detection module 400, wherein:

[0168] The acquisition module 100 is used to acquire the process data of the cutting machine within the first time range;

[0169] The prediction module 200 is used to input the process data into a pre-trained torque prediction model to obtain torque prediction data within a second time range.

[0170] The measurement module 300 is used to acquire the measured torque data of the cutting machine within a second time range;

[0171] The detection module 400 is used to determine the abnormal detection result of the spindle torque of the cutting machine based on the torque prediction data and the actual torque measurement data.

[0172] In one embodiment, the acquisition module 100 is further configured to:

[0173] Retrieve the first process data within a preset time range prior to the current moment stored in the database;

[0174] Obtain the second process data collected at the current moment;

[0175] Based on the data from the first process and the data from the second process, the process data is obtained.

[0176] In one embodiment, the cutting machine spindle torque anomaly detection device further includes a training module, which is used for:

[0177] Obtain historical process data of the cutting machine within a preset historical time range;

[0178] The process data is preprocessed to obtain historical process data;

[0179] Historical process data is used as training samples and input into a machine learning model for training to obtain a torque prediction model.

[0180] In one embodiment, the training module is also used for:

[0181] Time alignment is performed on each type of parameter in the historical process data to obtain time-aligned process data;

[0182] By filtering out abnormal data from the time-aligned process data, steady-state process data is obtained.

[0183] Based on preset feature selection rules, correlation analysis is performed on each type of parameter in the steady-state process data to obtain historical process data.

[0184] In one embodiment, the historical process data includes multiple cutting start positions, and the training module is also used for:

[0185] Based on the cutting start position, the time-aligned process data is batch-grouped to obtain multiple batches of cutting data;

[0186] Based on the multiple batches of cutting data, identify abnormal working conditions and uniform cutting time ranges in the batches of cutting data;

[0187] Based on the abnormal working condition and the uniform cutting time range, the cutting data of multiple batches were filtered to obtain steady-state process data.

[0188] In one embodiment, the training module is also used for:

[0189] The steady-state process data is processed by absolute value processing and smoothing filtering to obtain the processed steady-state process data.

[0190] In one embodiment, the machine learning model is an extreme gradient boosting model, and the training module is further used for:

[0191] The particle swarm for this extreme gradient boosting model is configured based on a multi-objective particle optimization algorithm;

[0192] Based on the historical process data and the extreme gradient model, the prediction results are obtained;

[0193] Based on the prediction results, the historical process data, and the preset loss function, the particle swarm optimization of the extreme gradient boosting model is adjusted.

[0194] The optimal position of the particle swarm is calculated based on a multi-objective particle swarm optimization algorithm.

[0195] The torque prediction model is determined based on the optimal position of the particle swarm obtained through multiple iterations of training.

[0196] In one embodiment, the detection module is further configured to:

[0197] Calculate the residual between the predicted torque data and the measured torque data;

[0198] If the number of data in the predicted torque data and the measured torque data whose residuals are greater than the residual threshold is greater than the preset number threshold, or the proportion of data whose residuals are greater than the residual threshold is greater than the preset proportion threshold, then the abnormal detection result of the spindle torque is abnormal.

[0199] Each module in the aforementioned cutting machine spindle torque anomaly detection device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of a computer device in hardware form or independent of it, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0200] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 9 As shown, the computer device includes a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When executed by the processor, the computer program implements a method for detecting abnormal spindle torque in a cutting machine. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0201] Those skilled in the art will understand that Figure 9 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0202] In one embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the cutting machine spindle torque abnormality detection method of any of the above embodiments.

[0203] Obtain the cutting machine's process data within the first time frame;

[0204] The process data is input into a pre-trained torque prediction model to obtain torque prediction data for the second time range;

[0205] Obtain the measured torque data of the cutting machine within the second time range;

[0206] Based on the predicted torque data and the measured torque data, the abnormal detection results of the spindle torque of the cutting machine are determined.

[0207] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the cutting machine spindle torque anomaly detection method of any of the above embodiments:

[0208] Obtain the cutting machine's process data within the first time frame;

[0209] The process data is input into a pre-trained torque prediction model to obtain torque prediction data for the second time range;

[0210] Obtain the measured torque data of the cutting machine within the second time range;

[0211] Based on the predicted torque data and the measured torque data, the abnormal detection results of the spindle torque of the cutting machine are determined.

[0212] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0213] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0214] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0215] The above embodiments merely illustrate several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for detecting abnormal spindle torque in a cutting machine, characterized in that, The method for detecting abnormal spindle torque of the cutting machine includes: Obtain the cutting machine's process data within the first time frame; The process data is input into a pre-trained torque prediction model to obtain torque prediction data within a second time range; Obtain the measured torque data of the cutting machine within the second time range; Based on the torque prediction data and the measured torque data, the abnormal detection result of the spindle torque of the cutting machine is determined.

2. The method for detecting abnormal spindle torque of a cutting machine according to claim 1, characterized in that, The acquisition of the cutting machine's process data within the first time range includes: Retrieve the first process data within a preset time range prior to the current moment stored in the database; Obtain the second process data collected at the current moment; The process data is obtained based on the first process data and the second process data.

3. The method for detecting abnormal spindle torque of a cutting machine according to claim 1, characterized in that, Before inputting the process data into the pre-trained torque prediction model to obtain torque prediction data within the second time range, the process further includes: Obtain historical process data of the cutting machine within a preset historical time range; The process data is preprocessed to obtain historical process data; Historical process data is used as training samples and input into a machine learning model for training to obtain a torque prediction model.

4. The method for detecting abnormal spindle torque of a cutting machine according to claim 3, characterized in that, The preprocessing of the historical process data to obtain historical process data includes: Time alignment is performed on each type of parameter in the historical process data to obtain time-aligned process data; The time-aligned process data is filtered for outliers to obtain steady-state process data; Based on preset feature selection rules, correlation analysis is performed on each type of parameter in the steady-state process data to obtain historical process data.

5. The method for detecting abnormal spindle torque of a cutting machine according to claim 4, characterized in that, The historical process data includes multiple cutting start positions. Data filtering of the time-aligned process data yields steady-state process data, including: Based on the cutting start position, the time-aligned process data is batch-grouped to obtain multiple batches of cutting data; Based on the multiple batches of cutting data, identify abnormal working conditions and uniform cutting time ranges in the batches of cutting data; Based on the abnormal operating conditions and the uniform cutting time range, multiple batches of cutting data are filtered to obtain steady-state process data.

6. The method for detecting abnormal spindle torque of a cutting machine according to claim 4, characterized in that, Before performing correlation analysis on each type of parameter in the steady-state process data based on preset feature selection rules to obtain historical process data, the method further includes: The steady-state process data is subjected to absolute value processing and smoothing filtering to obtain the processed steady-state process data.

7. The method for detecting abnormal spindle torque of a cutting machine according to claim 3, characterized in that, The machine learning model is an extreme gradient boosting model. The step of using historical process data as training samples and inputting them into the machine learning model for training to obtain the torque prediction model further includes: The particle swarm of the extreme gradient boosting model is configured based on a multi-objective particle optimization algorithm; Based on the historical process data and the extreme gradient model, the prediction results are obtained; Based on the prediction results, the historical process data, and the preset loss function, the particle swarm optimization of the extreme gradient boosting model is adjusted. The optimal position of the particle swarm is calculated based on a multi-objective particle swarm optimization algorithm. The torque prediction model is determined based on the optimal position of the particle swarm obtained through multiple iterations of training.

8. The method for detecting abnormal spindle torque of a cutting machine according to claim 1, characterized in that, The determination of the abnormal detection result of the spindle torque of the cutting machine based on the torque prediction data and the measured torque data includes: Calculate the residual between the predicted torque data and the measured torque data; If the number of data in the predicted torque data and the measured torque data whose residuals are greater than the residual threshold is greater than a preset number threshold, or the proportion of data whose residuals are greater than the residual threshold is greater than a preset proportion threshold, then the abnormal detection result of the spindle torque is abnormal.

9. A device for detecting abnormal spindle torque of a cutting machine, characterized in that, The device includes: The acquisition module is used to acquire the process data of the cutting machine within the first time range; The prediction module is used to input the process data into a pre-trained torque prediction model to obtain torque prediction data within a second time range. The measurement module is used to acquire the measured torque data of the cutting machine within a second time range; The detection module is used to determine the abnormal detection result of the spindle torque of the cutting machine based on the torque prediction data and the measured torque data.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 7.

Images