An abnormality detection method, device and storage medium of a tool changer speed reducer

Abstract

The application discloses a tool magazine tool changing speed reducer abnormality detection method and device and a storage medium, relates to the tool abnormality detection technical field, and comprises the following steps: in the machine tool machining process, the high-frequency operation state parameter set of the tool changing speed reducer is monitored and acquired, and a coarse abnormality risk coefficient sequence is output; the operation state parameter set of the tool changing speed reducer is monitored and acquired according to a second acquisition frequency; if the coarse abnormality risk coefficient sequence meets a first preset early warning index, a fine abnormality risk coefficient sequence is output; and if the fine abnormality risk coefficient sequence meets a second preset early warning index, the tool changing speed reducer is subjected to abnormality early warning. The application solves the technical problems that the existing technology cannot realize graded early warning for tool changing speed reducer abnormality state detection, resulting in insufficient sensitivity and low reliability, achieves graded early warning for tool changing speed reducer abnormality state, and improves the technical effects of detection sensitivity and reliability.

Description

Technical Field

[0001] This invention relates to the field of tool anomaly detection technology, specifically to an anomaly detection method, equipment, and storage medium for a tool magazine changer reducer. Background Technology

[0002] In the field of machine tool processing, the tool changer reducer is a key transmission component, and its abnormal condition can lead to decreased machining accuracy or even equipment failure. Existing detection methods generally suffer from fixed monitoring frequencies and single-level risk identification, making it difficult to accurately capture abnormal trends in the early stages. A single acquisition frequency cannot balance monitoring efficiency and accuracy, and fixed threshold early warning mechanisms are prone to false alarms or missed alarms due to noise interference. Furthermore, there is a lack of a tiered detection mechanism to achieve a progressive judgment from initial risk screening to accurate confirmation.

[0003] Existing technologies have the technical problem of being unable to provide graded early warning for abnormal conditions of tool changer reducers, resulting in insufficient sensitivity and low reliability. Summary of the Invention

[0004] This application provides a method, device, and storage medium for detecting abnormalities in a tool changer reducer, which addresses the technical problem in the prior art where the detection of abnormal states in the tool changer reducer cannot achieve graded early warning, resulting in insufficient sensitivity and low reliability.

[0005] In view of the above problems, this application provides a method, device and storage medium for abnormal detection of a tool changer reducer.

[0006] A first aspect of this application provides a method for detecting anomalies in a tool changer reducer, the method comprising:

[0007] During machine tool processing, a set of high-frequency operating status parameters of the tool changer is acquired through fixed-point monitoring at a first acquisition frequency. A small anomaly detection model is used for risk identification, and a coarse anomaly risk coefficient sequence is output. A set of operating status parameters of the tool changer is then acquired through fixed-point monitoring at a second acquisition frequency. If the coarse anomaly risk coefficient sequence meets a first preset early warning indicator, a large anomaly detection model is used to identify risks in the operating status parameter set, and a fine anomaly risk coefficient sequence is output. If the fine anomaly risk coefficient sequence meets a second preset early warning indicator, an anomaly warning is issued for the tool changer.

[0008] In a second aspect, this application provides an electronic device comprising: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to execute an anomaly detection method for a tool changer reducer provided in this application.

[0009] In a third aspect of this application, a computer-readable storage medium is provided, storing a computer program for executing an anomaly detection method for a tool changer reducer provided in this application.

[0010] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0011] During machine tool processing, a set of high-frequency operating status parameters of the tool changer is acquired through fixed-point monitoring at a first acquisition frequency, and a coarse anomaly risk coefficient sequence is output. A set of operating status parameters of the tool changer is then acquired through fixed-point monitoring at a second acquisition frequency. If the coarse anomaly risk coefficient sequence meets a first preset early warning indicator, a fine anomaly risk coefficient sequence is output. If the fine anomaly risk coefficient sequence meets a second preset early warning indicator, an anomaly warning is issued for the tool changer. This achieves graded early warning of abnormal states of the tool changer, improving the technical effect of detection sensitivity and reliability. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 A flowchart illustrating an abnormality detection method for a tool changer reducer provided in this application embodiment;

[0014] Figure 2 This is a schematic diagram of the structure of an electronic device provided in this application.

[0015] Explanation of reference numerals in the attached drawings: Processor 21, Memory 22, Input device 23, Output device 24. Detailed Implementation

[0016] This application provides a method, device, and storage medium for detecting abnormalities in a tool changer reducer, which addresses the technical problem in the prior art where the detection of abnormal states in the tool changer reducer cannot achieve graded early warning, resulting in insufficient sensitivity and low reliability.

[0017] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0018] Example 1, as Figure 1 As shown, this application provides a method for detecting abnormalities in a tool changer reducer, the method comprising:

[0019] Step S100: During the machine tool processing, the high-frequency operating status parameter set of the tool changer is obtained by fixed-point monitoring according to the first acquisition frequency. Risk identification is performed through the anomaly detection small model, and a coarse anomaly risk coefficient sequence is output.

[0020] Specifically, during machine tool processing, the preset operating status monitoring parameters of the tool changer reducer, including several initial monitoring indicators, are first acquired. Through correlation analysis, a preset proportion (less than 30%) of the indicators with the highest correlation are selected from the initial monitoring indicators as high-frequency monitoring indicators, forming a high-frequency monitoring indicator set. This set of high-frequency operating status parameters is then acquired through fixed-point monitoring at a relatively high initial acquisition frequency. Next, using the high-frequency monitoring indicator set as a constraint, multiple sample high-frequency monitoring indicator sets are collected based on historical operation and maintenance monitoring records. The failure probability under each sample state is statistically analyzed to generate a sample anomaly risk coefficient set. This set is used as training data to train a generative adversarial network until convergence, resulting in a small anomaly detection model. Because this small model only targets high-frequency monitoring indicators (with fewer indicators), it has high recognition efficiency but low accuracy. Finally, the sequence of high-frequency operating status parameters within a preset time window is input into the small anomaly detection model to complete coarse-grained risk identification, outputting a coarse anomaly risk coefficient sequence reflecting potential anomaly risks. This provides a preliminary screening basis for subsequent high-precision identification using a larger model with complete indicators.

[0021] Step S200: Obtain the set of operating status parameters of the tool changer reducer by fixed-point monitoring according to the second acquisition frequency.

[0022] Specifically, based on all initial monitoring indicators (including those not selected as high-frequency monitoring indicators), the tool changer reducer is monitored at a second acquisition frequency lower than the first acquisition frequency to obtain a complete set of operating status parameters. Simultaneously, a preset time period for risk trend judgment is configured, and the coarse anomaly risk coefficient sequence within this period is extracted and its trend analyzed to calculate the coarse anomaly risk growth rate. A frequency compensation coefficient is generated using formula 1 + coarse anomaly risk growth rate × preset constant (≤15), and the second acquisition frequency is periodically and dynamically adjusted to automatically increase the acquisition frequency when the risk rises, enhancing the monitoring sensitivity to abnormal states and achieving reasonable allocation of monitoring resources and dynamic adaptation to risk changes.

[0023] Step S300: If the coarse anomaly risk coefficient sequence meets the first preset early warning index, the operation status parameter set is risk-identified through the anomaly detection big model, and the fine anomaly risk coefficient sequence is output.

[0024] Specifically, if the coarse anomaly risk coefficient sequence satisfies a first preset early warning indicator including a coarse risk coefficient threshold and a coarse risk frequency threshold (where the coarse risk coefficient threshold is less than the subsequent fine risk coefficient threshold, and the coarse risk frequency threshold is less than the fine risk frequency threshold), then a large-scale anomaly detection model is constructed based on all initial monitoring indicators. The specific process is as follows: using the initial monitoring indicators as constraints, multiple sample monitoring indicator sets and sample anomaly risk coefficient sets are collected as training data. The data is divided into K equal parts (10≤K≤30), and K training sets are obtained by selecting with replacement K times. The generative adversarial network is then trained under supervision until convergence, resulting in K large anomaly detection branches, and a large-scale anomaly detection model is constructed. Subsequently, the average overflow value of the coarse risk coefficient sequence when it meets the first preset warning index is calculated. The number of branches to be selected is obtained by rounding down (average overflow value of coarse risk coefficient ÷ average overflow value of historical maximum coarse risk coefficient) × K. P branches are randomly selected from K large branches of anomaly detection. Risk identification is performed on the acquired sequence of operating status parameters (based on the collection of all initial monitoring indicators), and a more accurate fine anomaly risk coefficient sequence is output, realizing progressive detection from coarse-grained warning to fine-grained risk identification.

[0025] Step S400: If the precision anomaly risk coefficient sequence meets the second preset early warning index, then an anomaly warning is issued for the tool changer reducer.

[0026] Specifically, if the sequence of precise anomaly risk coefficients meets the second preset early warning index, which includes a precise risk coefficient threshold and a precise risk frequency threshold (where the precise risk coefficient threshold is higher than the coarse risk coefficient threshold in the first preset early warning index, and the precise risk frequency threshold is higher than the coarse risk frequency threshold), it is determined that there is a real anomaly risk in the tool changer reducer, and the anomaly early warning mechanism is triggered. This early warning is sent to maintenance personnel via indicator lights, pop-ups, or SMS messages to accurately locate the abnormal state, enabling timely intervention and maintenance to avoid machine tool downtime or machining quality problems caused by reducer failure, thus achieving closed-loop management and proactive prevention of equipment operation risks.

[0027] In one possible implementation, step S100 further includes:

[0028] Step S110: Obtain the preset operating status monitoring parameters of the tool changer reducer, wherein the preset operating status monitoring parameters include several initial monitoring indicators.

[0029] Step S120: Perform correlation analysis on the several initial monitoring indicators and the abnormal state of the tool changer reducer respectively, and select the initial monitoring indicator with the maximum correlation of the preset proportion as the high-frequency monitoring indicator to obtain the high-frequency monitoring indicator set, wherein the preset proportion is less than 30%.

[0030] Step S130: Based on the high-frequency monitoring index set, obtain the high-frequency operating status parameter set of the tool changer by fixed-point monitoring according to the first acquisition frequency.

[0031] Specifically, preset operating status monitoring parameters for the tool changer reducer are acquired. These parameters cover multiple initial monitoring indicators, such as tool change time (normal range 0.8s-1.5s, used to identify faults such as jamming and poor deceleration); peak motor current (normal range 1.5A-3.0A, used to check if the tool changer load has increased abnormally); current fluctuation range ±0.3A, an indicator of load stability; current duration is related to tool change time, abnormal duration indicates increased frictional resistance during tool change; output torque (normal range 1.2-2.0Nm, increased torque may indicate insufficient lubrication); root mean square vibration (RMS) value (normal range 0.1-0.5g, reflecting loose assembly); spectral characteristic peaks are used to detect local gear or bearing damage; temperature rise rate less than 2℃ / min indicates lubrication status; tool changer angular displacement accurate to ±0.5° of the specified tool position indicates positional deviation. These indicators together constitute the preset operating status monitoring parameters, providing basic data for subsequent analysis.

[0032] For several initial monitoring indicators of the tool changer reducer (such as tool change time and peak motor current), methods such as Pearson correlation coefficient are used to analyze their correlation with abnormal conditions of the tool changer reducer (such as jamming and insufficient lubrication). By quantifying the correlation values, the correlations are sorted from high to low, and the initial monitoring indicators with the highest correlation are selected according to a preset proportion of less than 30%. For example, if there are 10 initial monitoring indicators, a maximum of 3 indicators with the highest correlation are selected and set as high-frequency monitoring indicators, thus forming a high-frequency monitoring indicator set. This operation aims to focus on key indicators closely related to equipment anomalies, laying the foundation for efficient monitoring of equipment operating status and accurate identification of potential risks.

[0033] Based on the established high-frequency monitoring indicator set (which is a set of key indicators selected from the initial monitoring indicators that have the highest correlation with the abnormal state of the tool changer and account for less than 30%), a relatively high initial acquisition frequency is used for fixed-point monitoring (this frequency setting is intended to achieve high-frequency monitoring of key indicators to capture subtle changes in equipment operation). During the operation of the tool changer, real-time data is collected from various indicators in the high-frequency monitoring indicator set (such as tool change time, motor current peak, and other key indicators) at fixed time intervals or under specific trigger conditions. The collected indicator data are integrated to form a high-frequency operating status parameter set for the tool changer. This parameter set contains key operating status information of the equipment, providing data support for subsequent preliminary risk identification through anomaly detection models.

[0034] In one possible implementation, step S100 further includes:

[0035] Step S140: Using the high-frequency monitoring index set as a constraint, collect multiple sample high-frequency monitoring index sets based on the historical operation and maintenance monitoring records of the tool changer reducer, calculate the failure probability of the tool changer reducer under different sample high-frequency monitoring index set states, set it as the sample abnormality risk coefficient, and obtain the sample abnormality risk coefficient set.

[0036] Step S150: Using the multiple sets of high-frequency monitoring indicators and the set of sample anomaly risk coefficients as training data, train the generative adversarial network until convergence to obtain a small anomaly detection model.

[0037] Step S160: Monitor and obtain the high-frequency operating status parameter set sequence within the preset time window, input the anomaly detection small model at a fixed point for risk identification, and output the coarse anomaly risk coefficient sequence.

[0038] Specifically, using a defined set of high-frequency monitoring indicators as constraints, multiple sample sets of high-frequency monitoring indicators are selectively collected from the historical maintenance monitoring records accumulated over a long period of time for the tool changer reducer. These samples cover the equipment's operational data under different operating conditions and at different time points. For each sample set of high-frequency monitoring indicators, its corresponding operating status is analyzed in detail. By studying the actual occurrence of equipment failures and combining information such as historical maintenance records and failure reports, the probability of the tool changer reducer failing under that sample status is statistically determined. This probability value is set as the sample anomaly risk coefficient. By summing up the anomaly risk coefficients corresponding to multiple samples, a sample anomaly risk coefficient set is formed. This set is a quantitative representation of the equipment's historical operational risk status, providing crucial foundational data for the subsequent training of the generative adversarial network, enabling the trained anomaly detection mini-model to learn the correlation pattern between equipment operating status and failure risk.

[0039] Multiple sets of high-frequency monitoring indicators and their corresponding anomaly risk coefficients are used as training data and input into a Generative Adversarial Network (GAN) for supervised training. The GAN continuously optimizes its parameters through a game process where the generator fits the distribution of the sample data, and the discriminator distinguishes between real and generated data. Specifically, during training, the high-frequency monitoring indicator sets are used as input features, and the anomaly risk coefficients are used as labels. The network weights are adjusted using a backpropagation algorithm until the distribution of simulated data output by the generator closely approximates the distribution of real samples, and the discriminator cannot accurately distinguish between real and fake data; at this point, the network is considered to have converged. The resulting small anomaly detection model has the ability to perform coarse-grained risk identification of the tool changer's operating status based on the high-frequency monitoring indicator set. It can quickly output a coarse anomaly risk coefficient sequence reflecting potential anomalies through lightweight computation.

[0040] According to the first acquisition frequency, the high-frequency operating status parameters of the tool changer reducer within a preset time window (e.g., the most recent 10 minutes) are continuously monitored, forming a parameter set sequence containing multiple sets of high-frequency monitoring index data (e.g., acquiring high-frequency monitoring indicators once per second, forming a sequence of 600 data sets). Subsequently, this parameter set sequence is input into a trained anomaly detection mini-model at fixed time intervals (i.e., fixed points). Since this model is trained only on the high-frequency monitoring index set (accounting for less than 30% of the initial indexes), it possesses lightweight computational characteristics and can quickly perform coarse-grained risk identification on the input data. The model outputs risk probability values ​​through the discriminator of a generative adversarial network, which are then combined with historical thresholds to map to corresponding coarse anomaly risk coefficients. Finally, a coarse anomaly risk coefficient sequence reflecting the trend of equipment operating risks is generated in chronological order, providing a preliminary basis for subsequent judgment on whether to trigger fine detection by the large model.

[0041] In one possible implementation, step S200 further includes:

[0042] Step S210: Based on the aforementioned initial monitoring indicators, the operating status parameter set of the tool changer is obtained by fixed-point monitoring according to the second acquisition frequency, wherein the second acquisition frequency is less than the first acquisition frequency.

[0043] Step S220: Configure the preset time period for risk trend judgment, obtain the coarse anomaly risk coefficient sequence within the preset time period, perform coarse anomaly risk trend analysis, and output the coarse anomaly risk growth rate.

[0044] Step S230: Use 1 plus the product of the crude abnormal risk growth rate and the preset constant as the frequency compensation coefficient, wherein the preset constant is less than or equal to 15.

[0045] Step S240: Adjust the second acquisition frequency periodically according to the frequency compensation coefficient.

[0046] Specifically, based on all determined initial monitoring indicators (covering a complete parameter system including vibration frequency, temperature, and rotational speed), the tool changer reducer is monitored at a second sampling frequency lower than the first sampling frequency (e.g., the first sampling frequency is 10 times per second, and the second sampling frequency is 1 time per minute) to obtain a set of operating status parameters containing all initial indicators. Through differentiated frequency design, while ensuring overall equipment status coverage, the monitoring resource consumption of non-critical indicators is reduced, forming a hierarchical monitoring mechanism that focuses on key indicators at high frequencies and covers complete parameters at low frequencies. This lays the foundation for subsequent dynamic adjustment of monitoring strategies in conjunction with coarse-grained anomaly risks.

[0047] A preset time period (e.g., configurable time intervals such as 30 minutes or 1 hour) is set to assess risk change trends. The coarse anomaly risk coefficient sequence within this period is extracted from stored historical data. Time series analysis methods such as moving average, exponential smoothing, or linear regression are used to quantitatively analyze the changing trends of the coarse anomaly risk coefficients, calculating the growth rate of the risk coefficient per unit time and outputting the coarse anomaly risk growth rate. This growth rate reflects the evolution of the tool changer reducer's abnormal risk over time, providing crucial trend judgment basis for subsequent dynamic adjustment of monitoring frequency and triggering of fine detection mechanisms.

[0048] The frequency compensation coefficient is calculated using formula 1 + coarse anomaly risk growth rate × preset constant, where the preset constant is a configurable parameter with a value ≤ 15. The design logic of this formula is as follows: when the coarse anomaly risk growth rate is positive, it indicates an upward trend in equipment anomaly risk. By introducing the preset constant, the influence weight of the growth rate is amplified, making the compensation coefficient greater than 1, thereby increasing the second acquisition frequency. If the growth rate is negative or zero, the compensation coefficient is ≤ 1, maintaining or reducing the acquisition frequency. For example, if the coarse anomaly risk growth rate is 0.2 / hour and the preset constant is 10, the frequency compensation coefficient is 1 + 0.2 × 10 = 3. In this case, the second acquisition frequency will be adjusted to three times the original frequency, thus achieving a dynamic response to risk changes, increasing monitoring density when risks intensify, and optimizing resource allocation efficiency.

[0049] The mechanism for periodically adjusting the second acquisition frequency based on the frequency compensation coefficient identifies the abnormal trend of the coarse anomaly risk coefficient sequence and uses a time series analysis algorithm to fit the coarse anomaly risk growth rate. Based on this growth rate, the frequency compensation coefficient is dynamically calculated using 1 + coarse anomaly risk growth rate × a preset constant. This allows for adjustment of the acquisition frequency (i.e., the second acquisition frequency) for all types of data based on all initial monitoring indicators. When the equipment is operating stably and the risk growth rate is low, maintaining a lower second acquisition frequency saves acquisition bandwidth, processing power, and storage space, reducing operation and maintenance costs. Conversely, when a rapid upward trend in risk is detected (i.e., a high coarse anomaly risk growth rate), the frequency compensation coefficient increases accordingly, and the second acquisition frequency automatically increases proportionally, thereby obtaining more refined equipment operating status data. This mechanism enables refined dynamic monitoring of the gearbox status. On the one hand, it optimizes the allocation of monitoring resources through a differentiated frequency strategy, ensuring comprehensive monitoring while avoiding resource waste. On the other hand, through risk trend-driven dynamic parameter adjustment, it significantly enhances the timeliness and sensitivity of anomaly detection, improves monitoring accuracy in the early stages of a fault, triggers early warnings, buys more intervention time for maintenance personnel, effectively avoids serious equipment damage, and ultimately improves the flexibility and reliability of predictive maintenance.

[0050] In one possible implementation, step S300 further includes:

[0051] Step S310: Configure a first preset early warning indicator and a second preset early warning indicator, wherein the first preset early warning indicator includes a coarse risk coefficient threshold and a coarse risk frequency threshold, and the second preset early warning indicator includes a fine risk coefficient threshold and a fine risk frequency threshold, wherein the coarse risk coefficient threshold is less than the fine risk coefficient threshold, and the coarse risk frequency threshold is less than the fine risk frequency threshold.

[0052] Specifically, a first and second preset early warning indicator need to be configured to construct a hierarchical anomaly detection system. The first preset early warning indicator serves as a preliminary screening standard, including a coarse risk coefficient threshold and a coarse risk frequency threshold. The coarse risk coefficient threshold is set to a relatively low risk probability value (e.g., 0.3) to cover a wider range of potential anomalies and avoid missed detections. The coarse risk frequency threshold is defined as the number of consecutive or cumulative exceedances of the coarse risk coefficient threshold (e.g., 3 consecutive exceedances or 5 exceedances in 10 monitoring sessions), used to exclude interference from random fluctuations. The second preset early warning indicator serves as a precise early warning standard, including a precise risk coefficient threshold and a precise risk frequency threshold. The precise risk coefficient threshold is higher than the coarse risk coefficient threshold (e.g., 0.6), requiring a higher risk confidence level; the precise risk frequency threshold is also higher than the coarse risk frequency threshold (e.g., 5 consecutive exceedances or 7 exceedances in 10 monitoring sessions) to ensure the rigor of anomaly detection. By employing a tiered design with a small coarse threshold and a wide coverage, and a high fine threshold and strict judgment criteria, a progressive early warning system is achieved, from initial risk identification to accurate anomaly confirmation. This ensures sensitive capture of early risks while avoiding false alarms or missed alarms caused by a single threshold, thereby improving the reliability and practicality of the detection results.

[0053] In one possible implementation, step S300 further includes:

[0054] Step S320: Construct a large-scale anomaly detection model based on the aforementioned initial monitoring indicators.

[0055] Step S330: Monitor and acquire the sequence of operating status parameters within a preset time window, sequentially perform risk identification through the large anomaly detection model, and output the sequence of refined anomaly risk coefficients.

[0056] Specifically, a large-scale anomaly detection model is constructed based on the aforementioned initial monitoring indicators. The process is as follows: First, constrained by all initial monitoring indicators, multiple sample monitoring indicator sets containing complete indicators are collected from the historical operation and maintenance monitoring records of the tool changer reducer. Simultaneously, the corresponding sample anomaly risk coefficient set (reflecting the failure probability under different indicator combinations) is obtained, forming a training dataset. Then, the training data is divided into K equal parts (10≤K≤30), and K times are selected from the K parts using sampling with replacement to generate K training sets (each training set may contain duplicate data to enhance model robustness). For each training set, a Generative Adversarial Network (GAN) is trained under supervision. Through the game process of the generator fitting the data distribution and the discriminator distinguishing between real and generated data, the network parameters are continuously adjusted until each branch model converges, ultimately obtaining a large-scale anomaly detection model composed of K large anomaly detection branches. Because this model covers all initial monitoring indicators, it can perform full-dimensional feature extraction and risk analysis of equipment operating status, providing high-resolution detection capabilities for subsequent precise anomaly risk identification.

[0057] When the coarse anomaly risk coefficient sequence meets the first preset warning indicator, the operating status parameter sequence within a preset time window (e.g., the most recent 30 minutes) is first monitored and acquired. This sequence contains all the initial monitoring indicator data (covering the complete parameter system that was not selected as a high-frequency indicator). Subsequently, this parameter sequence is input into the constructed anomaly detection large model for risk identification. The specific process is as follows: First, the average overflow value of the coarse risk coefficient when the coarse anomaly risk coefficient sequence meets the first preset warning indicator (i.e., the average value exceeding the coarse risk coefficient threshold) is calculated. Then, the number of branches to be selected, P (K is the total number of branches in the large model, 10≤K≤30), is obtained by rounding down using the formula (average overflow value of coarse risk coefficient ÷ average overflow value of historical maximum coarse risk coefficient) × K. Then, P branches are randomly selected from the K large anomaly detection branches to perform parallel analysis on the input operating status parameter sequence. Each branch independently outputs a risk probability value based on the feature patterns learned from different training subsets, and finally, they are integrated to generate a refined anomaly risk coefficient sequence. This sequence achieves high-precision quantification of equipment anomaly risks through the input of full-indicator data and the collaborative judgment of multi-branch models. Compared with the coarse-grained detection of small models, it significantly improves the accuracy and reliability of risk identification, providing a key basis for triggering accurate early warnings in the future.

[0058] In one possible implementation, step S320 further includes:

[0059] Step S321: Using the aforementioned initial monitoring indicators as constraints, collect multiple sets of sample monitoring indicators and sets of sample anomaly risk coefficients based on historical operation and maintenance monitoring records.

[0060] Step S322: Use the multiple sample monitoring index sets and sample abnormal risk coefficient sets as training data, and divide them into K equal parts. Select the first training set K times with replacement from the K sets of data. Iterate and select K times to obtain K training sets, where K is an integer greater than or equal to 10 and less than or equal to 30.

[0061] Step S323: Using the K training sets, supervise the training of the generative adversarial network until convergence, and obtain K major branches for anomaly detection to build a large anomaly detection model.

[0062] Specifically, using the aforementioned initial monitoring indicators (a complete parameter system covering the operating status of the tool changer reducer) as constraints, multiple sample monitoring indicator sets are collected from the equipment's historical operation and maintenance monitoring records. These samples include actual monitoring data for all initial indicators such as vibration frequency, temperature, speed, and current. Simultaneously, for each sample monitoring indicator set, the failure probability of the tool changer reducer under that set of indicator data is statistically analyzed by analyzing historical fault records, and this probability is set as the sample anomaly risk coefficient, forming a sample anomaly risk coefficient set. The collected samples need to cover various operating conditions, including normal equipment operation, potential anomalies, and fault states, to ensure the diversity and representativeness of the training data. This provides real and comprehensive input data support for the subsequent construction of a large-scale anomaly detection model, enabling the model to learn the mapping relationship between different operating states and fault risks.

[0063] Multiple sets of collected sample monitoring indicators and their corresponding anomaly risk coefficients are used as training data. First, this data is evenly divided into K parts (K is an integer between 10 and 30), each containing a subset of samples and their risk coefficients. Then, sampling with replacement is used to randomly select K samples from these K parts to form the first training set. Each sampling allows for repeated selection of the same data set to increase the diversity of the training data and the robustness of the model. By iterating this sampling process K times, K different training sets are finally obtained, each containing the feature distribution of the original data but with different sample combinations. This bootstrap sampling method effectively utilizes limited historical data to generate multiple sets of statistically significant training samples, avoiding model overfitting due to insufficient data. It provides a rich dataset for subsequent multi-branch training of the generative adversarial network, ensuring that each branch of the large anomaly detection model can learn risk features of different dimensions.

[0064] For the obtained K training sets (K being an integer from 10 to 30), the Generative Adversarial Network (GAN) is independently trained using supervised training. Each training set contains a set of sample monitoring indicators and corresponding sample anomaly risk coefficients generated from the original data through sampling with replacement. These serve as input data to drive the generator and discriminator of the GAN in a game-like learning process: the generator is responsible for fitting the distribution of the input data to generate simulated data that closely approximates the real samples; the discriminator is used to distinguish between real data and generated data. The network parameters are continuously adjusted through backpropagation until the data distribution output by the generator is sufficiently close to the distribution of the real data, and the discriminator can no longer accurately distinguish between the two, i.e., the model is considered to have converged. After K independent training iterations, K large branches for anomaly detection with different feature extraction capabilities are finally generated. Each branch corresponds to a set of optimized network parameters, capable of capturing different risk feature patterns from the full set of initial monitoring indicators. These K branches are combined into a large anomaly detection model, enabling it to perform multi-branch parallel analysis. By integrating multi-dimensional features, the detection accuracy of complex abnormal states of the tool changer reducer can be improved, providing strong model support for the subsequent output of precise anomaly risk coefficients.

[0065] In one possible implementation, step S330 further includes:

[0066] Step S331: Calculate the average overflow value of the coarse risk coefficients when the coarse anomaly risk coefficient sequence meets the first preset early warning index.

[0067] Step S332: Calculate the ratio of the average overflow of the crude risk coefficient to the average overflow of the historical maximum crude risk coefficient, and multiply it by K and round it to obtain the number of branches selected, P.

[0068] Step S333: Within the K major anomaly detection branches of the anomaly detection model, randomly select P major anomaly detection branches, perform risk identification on the running state parameter sequence, and output a refined anomaly risk coefficient sequence.

[0069] Specifically, when the coarse anomaly risk coefficient sequence meets the first preset warning indicator (i.e., the coarse risk coefficient exceeds the coarse risk coefficient threshold and reaches the coarse risk frequency threshold), all risk coefficient data points in the sequence that exceed the coarse risk coefficient threshold are first selected. For each data point that meets the condition, the difference between it and the coarse risk coefficient threshold is calculated (i.e., the actual risk coefficient minus the threshold), resulting in an overflow value. Subsequently, the arithmetic mean of these overflow values ​​is calculated to obtain the coarse risk coefficient overflow mean. This mean quantifies the average degree to which the current anomaly risk exceeds the preliminary warning threshold, providing key parameters for subsequent assessment of risk severity and dynamic adjustment of the number of branches in the fine detection model, ensuring that a more refined detection process can be triggered when equipment risk increases.

[0070] The calculated average overflow of the coarse risk coefficient (i.e., the average degree to which the current coarse risk coefficient exceeds the threshold) is compared with the average maximum overflow of the coarse risk coefficient in historical records. The ratio between the two is calculated to reflect the relative level of the current risk overflow. This ratio is then multiplied by the total number of branches K (10 ≤ K ≤ 30) of the anomaly detection model, and rounded to obtain the number of branches selected, P. This process, through historical data normalization, allows the P value to dynamically respond to changes in risk. When the risk overflow level is high, the P value increases, and more model branches participate in fine detection; conversely, the number of participating branches decreases. Through this mechanism, computational resources can be intelligently allocated according to the severity of risk, optimizing computational efficiency while ensuring detection accuracy, and ensuring that the anomaly detection model can achieve efficient risk identification at different risk levels.

[0071] From the K major anomaly detection branches of the large anomaly detection model (K being an integer from 10 to 30), P branches are randomly selected according to the calculated number of branch selections P to participate in precise anomaly risk identification. The selection process employs a random strategy to ensure diversity in branch combinations for each detection, avoiding detection bias caused by fixed branches. For the sequence of operating status parameters within a preset time window (including all initial monitoring index data), the selected P branches perform risk identification in parallel, with each branch independently outputting a risk coefficient value based on the feature patterns learned from its training subset. Finally, the arithmetic mean of the outputs of the P branches is taken to generate a precise anomaly risk coefficient sequence. This multi-branch random selection and result fusion mechanism improves detection efficiency through parallel computation and enhances the reliability of detection results by utilizing the complementary features of different branches, enabling the precise anomaly risk coefficient sequence to more accurately reflect the true anomaly risk level of the equipment.

[0072] Example 2, Figure 2 This is a schematic diagram of the structure of an electronic device provided in Embodiment 2 of this application, and a block diagram of an exemplary electronic device suitable for implementing the embodiments of the present invention. Figure 2 The electronic device shown is merely an example and should not be construed as limiting the functionality or scope of the embodiments of the present invention. Figure 2 As shown, the electronic device includes a processor 21, a memory 22, an input device 23, and an output device 24; the number of processors 21 in the electronic device can be one or more. Figure 2 Taking a processor 21 as an example, the processor 21, memory 22, input device 23, and output device 24 in an electronic device can be connected via a bus or other means. Figure 2 Taking the example of a connection between China and Israel via a bus.

[0073] In embodiment three, the memory 22, as a computer-readable storage medium, can be used to store software programs, computer-executable programs, and modules, such as the program instructions / modules corresponding to the abnormal detection method of the tool changer reducer in this embodiment of the application. The processor 21 executes various functional applications and data processing of the computer device by running the software programs, instructions, and modules stored in the memory 22, thereby realizing the above-mentioned abnormal detection method of the tool changer reducer.

[0074] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0075] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0076] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application intends to include such modifications and variations.

Claims

1. A method for detecting abnormalities in a tool changer reducer, characterized in that, The methods include: During the machine tool processing, the high-frequency operating status parameter set of the tool changer is obtained by fixed-point monitoring according to the first acquisition frequency. Risk identification is performed through an anomaly detection small model, and a coarse anomaly risk coefficient sequence is output. The operating status parameter set of the tool changer reducer is obtained by fixed-point monitoring according to the second acquisition frequency; If the coarse anomaly risk coefficient sequence meets the first preset early warning index, the anomaly detection big model is used to identify the risk of the operating status parameter set and output the fine anomaly risk coefficient sequence. If the precision anomaly risk coefficient sequence meets the second preset early warning index, then an anomaly warning is issued for the tool changer reducer; The high-frequency operating status parameter set of the tool changer obtained by fixed-point monitoring according to the first acquisition frequency includes: Acquire preset operating status monitoring parameters for the tool changer reducer, wherein the preset operating status monitoring parameters include several initial monitoring indicators; The set of operating status parameters of the tool changer obtained by fixed-point monitoring according to the second acquisition frequency includes: Based on the aforementioned initial monitoring indicators, the operating status parameter set of the tool changer reducer is obtained by fixed-point monitoring at a second acquisition frequency, wherein the second acquisition frequency is less than the first acquisition frequency; Configure a preset time period for risk trend judgment, obtain the coarse anomaly risk coefficient sequence within the preset time period, perform coarse anomaly risk trend analysis, and output the coarse anomaly risk growth rate. The product of 1 and the gross anomaly risk growth rate and a preset constant is used as the frequency compensation coefficient, wherein the preset constant is less than or equal to 15; The second acquisition frequency is periodically adjusted according to the frequency compensation coefficient.

2. The method for detecting abnormalities in a tool changer reducer according to claim 1, characterized in that, The high-frequency operating status parameter set of the tool changer is obtained by fixed-point monitoring according to the first acquisition frequency, including: Correlation analysis was performed on the initial monitoring indicators and the abnormal state of the tool changer reducer, and the initial monitoring indicator with the highest correlation of a preset proportion was selected as the high-frequency monitoring indicator to obtain a high-frequency monitoring indicator set, wherein the preset proportion is less than 30%. Based on the high-frequency monitoring index set, the high-frequency operating status parameter set of the tool changer is obtained by fixed-point monitoring according to the first acquisition frequency.

3. The method for detecting abnormalities in a tool changer reducer according to claim 2, characterized in that, Risk identification is performed using a small anomaly detection model, outputting a coarse anomaly risk coefficient sequence, including: Using the high-frequency monitoring index set as a constraint, based on the historical operation and maintenance monitoring records of the tool changer reducer, multiple sample high-frequency monitoring index sets are collected, and the failure probability of the tool changer reducer under different sample high-frequency monitoring index set states is statistically analyzed and set as the sample abnormality risk coefficient, thereby obtaining the sample abnormality risk coefficient set. Using the high-frequency monitoring index set and the sample anomaly risk coefficient set of the multiple samples as training data, a generative adversarial network is trained until convergence to obtain a small anomaly detection model. The system monitors and acquires a set of high-frequency operating status parameters within a preset time window, inputs them into the anomaly detection mini-model for risk identification, and outputs a coarse anomaly risk coefficient sequence.

4. The method for detecting abnormalities in a tool changer reducer according to claim 1, characterized in that, Configure a first preset early warning indicator and a second preset early warning indicator, wherein the first preset early warning indicator includes a coarse risk coefficient threshold and a coarse risk frequency threshold, and the second preset early warning indicator includes a fine risk coefficient threshold and a fine risk frequency threshold, wherein the coarse risk coefficient threshold is less than the fine risk coefficient threshold, and the coarse risk frequency threshold is less than the fine risk frequency threshold.

5. The method for detecting abnormalities in a tool changer reducer according to claim 2, characterized in that, The operational state parameter set is risk-identified using a large-scale anomaly detection model, and a refined anomaly risk coefficient sequence is output, including: A large-scale anomaly detection model is constructed based on the aforementioned initial monitoring indicators; The system monitors and acquires the sequence of operating status parameters within a preset time window, sequentially identifies risks through a large anomaly detection model, and outputs a sequence of precise anomaly risk coefficients.

6. The method for detecting abnormalities in a tool changer reducer according to claim 5, characterized in that, Based on the aforementioned initial monitoring indicators, a large-scale anomaly detection model is constructed, including: Based on the aforementioned initial monitoring indicators and historical operation and maintenance monitoring records, multiple sets of sample monitoring indicators and sets of sample anomaly risk coefficients are collected. The multiple sets of sample monitoring indicators and the set of sample abnormal risk coefficients are used as training data and divided into K equal parts. The first training set is obtained by selecting the K sets of data with replacement K times. The first training set is obtained by iteratively selecting K sets of data, where K is an integer greater than or equal to 10 and less than or equal to 30. Using the K training sets, the generative adversarial network is trained under supervision until convergence, resulting in K major branches for anomaly detection, which are then used to build a large anomaly detection model.

7. The method for detecting abnormalities in a tool changer reducer according to claim 6, characterized in that, The system monitors and acquires a sequence of operational status parameters within a preset time window, sequentially processes these parameters through a large-scale anomaly detection model for risk identification, and outputs a sequence of refined anomaly risk coefficients, including: Calculate the average overflow value of the coarse risk coefficients when the coarse anomaly risk coefficient sequence meets the first preset early warning index; Calculate the ratio of the average overflow of the crude risk coefficient to the average overflow of the historical maximum crude risk coefficient, and multiply it by K and round it to obtain the number of branches selected, P. Within the K major anomaly detection branches of the anomaly detection model, P major anomaly detection branches are randomly selected to perform risk identification on the sequence of operating state parameters and output a sequence of precise anomaly risk coefficients.

8. An electronic device, characterized in that, The electronic device includes: Memory, used to store executable instructions; The processor, when executing executable instructions stored in the memory, implements the abnormal detection method for the tool changer reducer according to any one of claims 1-7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements an anomaly detection method for a tool changer reducer as described in any one of claims 1-7.

Images