A method for detecting faults of an alternating current motor in a digital twin workshop by using software and hardware in cooperation

Abstract

This invention discloses a hardware-software collaborative fault detection method for AC motors in a digital twin workshop, comprising the following steps: using an array-based hardware initial inspection method to obtain initial fault detection results for the AC motor; sending the initial fault detection results to the digital twin workshop system via a main controller; when the digital twin workshop system receives the initial fault detection results, using a KNN agent-based re-inspection method to obtain fault detection results; based on the fault detection results, the digital twin workshop system generates fault judgment information; when the digital twin workshop system does not receive the initial fault detection results, using a KNN agent-based prediction method to detect fault symptoms and obtain fault prediction results; and based on the fault prediction results, repairing the AC motor and recording the actual fault conditions and fault information. For large-scale motor deployments in a digital twin workshop, this invention can shorten the time for fault detection and initial fault location.

Description

Technical Field

[0001] This invention relates to the field of motor fault detection, and in particular to a hardware and software collaborative detection method for AC motor faults in a digital twin workshop. Background Technology

[0002] With the continuous development of industrial automation technology and the increasing level of enterprise informatization, data acquisition equipment, processing equipment, testing equipment, and control equipment are being used more and more widely at the workshop level. However, these devices with different types of interfaces combine to form a very complex industrial network. In such a network environment, effective network management and fault monitoring of various devices are particularly important, because once a device in a part of the network fails, it will inevitably affect other connected devices that are still operating normally, and in more serious cases, it will lead to the shutdown and collapse of the entire network.

[0003] Against this backdrop, the Digital Twin Workshop (DTW) has emerged as an efficient means of systematic workshop management. It achieves data fusion and collaborative evolution between the physical and information worlds by constructing a digital model in a virtual information space that is mapped in real-time to the physical workshop, enabling data exchange and dynamic interaction.

[0004] However, in actual workshop production and the deployment of digital twin systems, traditional fault detection methods face significant bottlenecks at the hardware level. Currently widely adopted solutions are mostly based on an independent monitoring mode of "one machine, one sensor," meaning that each motor is equipped with a separate current transformer and corresponding acquisition unit. For example, in Chinese patent document CN120820846A, when three motors connected in parallel in a multi-motor system are controlled by an IPM module, the output of the circuit switching controller is electrically connected to these three motors through three current holders. The first, second, and third motors are each electrically connected to the output of the circuit switching controller through a current holder. While this method can achieve independent monitoring of each device, it requires a one-to-one correspondence between the tested motor and the sampling device. In scenarios involving large-scale motor deployments, the required number of sensors and acquisition devices is enormous, leading to a sharp increase in hardware costs, complex wiring, and extended installation and maintenance cycles. Furthermore, the lack of reuse and coordination of detection resources among similar motors results in hardware redundancy and resource waste, making it difficult to meet the practical requirements of digital twin workshops for monitoring systems that are low-cost, easy to deploy, and scalable. In addition, traditional solutions often require the deployment of multiple data acquisition devices to obtain multi-dimensional operating data from multiple motors, further exacerbating deployment cost pressures.

[0005] At the software and algorithm level, traditional data-driven fault diagnosis models also have significant limitations. These models typically rely on complex machine learning or deep learning networks. Their training process requires collecting a large number of balanced, high-quality fault samples and involves tedious feature engineering, network structure adjustment, and hyperparameter optimization, resulting in high costs for initial preparation and training. For example, in the technical solution of Chinese patent document CN117251782A, the processor performs fault detection by reading data from a time-series database. This solution actually performs machine learning-based fault detection on the data obtained for each tested motor. This full-data processing mode consumes a lot of computing power and affects the timeliness of detection. At the same time, this solution relies on naturally occurring fault events to accumulate training data. Features and labeled data can only be extracted when the motor fails or is about to fail, resulting in time-consuming and labor-intensive processes and poor initial detection results. In the early stages of deployment, it faces the "cold start" problem of sparse samples and class imbalance.

[0006] Furthermore, most existing models are permanently deployed once trained, lacking a mechanism for continuous learning and self-optimization during actual operation. As equipment operating time accumulates, operating conditions change, or components age, the original model may gradually deviate from the actual state, resulting in a decline in diagnostic accuracy and making it difficult to meet the technical requirements of long-term stability and adaptive evolution of fault detection systems in digital twin workshops. For example, although the technical solution in Chinese patent application CN119165349A establishes a feedback mechanism, allowing maintenance personnel or managers to confirm or correct diagnostic results, this method still requires manual confirmation of results and does not establish the model's self-updating capability.

[0007] Furthermore, existing technical solutions typically focus only on the detection and diagnosis of existing faults, failing to effectively utilize idle system computing power for proactive prediction of potential faults. In digital twin workshops, a large number of motors operate continuously for extended periods, and the failure to detect fault signs and issue early warnings increases the risk of unplanned downtime. Summary of the Invention

[0008] Purpose of the Invention: The purpose of this invention is to provide a hardware and software collaborative detection method for AC motor faults in digital twin workshops. By designing an array-style cross-deployment strategy, the hardware scale and cost are significantly reduced, and a KNN agent with continuous learning capabilities is introduced to achieve accurate fault location and type identification. At the same time, the idle computing power of the system is used to perform fault prediction based on the same KNN agent to achieve early warning of fault symptoms. Thus, while improving the economy and reliability of detection, it effectively enhances the proactive maintenance capability of the digital twin workshop, ensuring the continuity and efficiency of production.

[0009] Technical Solution: A hardware and software collaborative detection method for AC motor faults in a digital twin workshop, comprising the following steps:

[0010] An array-based hardware initial inspection method is used to obtain the initial fault location result of the AC motor; the initial fault location result is sent to the digital twin workshop system via the main controller as the initial fault inspection result;

[0011] When the digital twin workshop system receives the initial fault detection result, it uses the KNN agent re-inspection method to obtain the fault detection result; based on the fault detection result, the digital twin workshop system generates fault judgment information.

[0012] When the digital twin workshop system does not receive the initial fault detection results, the KNN intelligent agent prediction method is used to detect fault symptoms and obtain fault prediction results. Based on the fault prediction results, the AC motor is repaired and the actual fault situation and fault information are recorded.

[0013] Furthermore, the specific steps of the array-type hardware initial inspection method include:

[0014] Sa1: Collect information on all motors that need to be fault-detected in the digital twin workshop equipment, and encode the motors to be inspected by serial number to obtain a motor code table for all motors to be inspected.

[0015] Sa2, based on the motor coding table and array-type cross-strategy, deploys a current acquisition array to collect AC signals from motors running in the workshop and outputs DC signals.

[0016] Sa3, the main controller collects DC signals, analyzes the DC signals and performs initial fault location to obtain the fault motor coding range;

[0017] Sa4 sends the faulty motor coding range to the digital twin workshop system.

[0018] Furthermore, the AC electrical signals from the motors running in the workshop are obtained. The specific steps are as follows:

[0019] Sa21, based on the motor coding table, yields the number of motors, the number of current collectors required for deploying the current collector array, and the number of current transformers. One current collector corresponds to one current transformer. The formula for calculating the number of current collectors is as follows:

[0020] ,

[0021] In the formula, The number of current collectors required to deploy a current collector array for type A motors. The number of motors of type A;

[0022] Sa22 determines the current acquisition array deployment strategy based on an array-based cross-deployment approach, as follows:

[0023] Let the motor code in the motor coding table be MOTOR(n), where n is a consecutive positive integer. ;

[0024] right The current collectors are sequentially encoded to obtain a current collector array. The encoding of each current collector is represented by m, where m is a consecutive positive integer. ;

[0025] The specific relationship between the current acquisition device and the motor being acquired in the array-type cross-deployment strategy is as follows:

[0026] 1) The current collector collects the total current of all motors;

[0027] 2) When m is not equal to At that time, the m-th current collector collects the motor MOTOR(m) and the motor MOTOR(m). The total current of all motors between )

[0028] Sa23 measures the magnitude of the current collected by the current collector array and converts the AC current signal of the motor into a DC signal.

[0029] Furthermore, the specific steps to obtain the faulty motor coding range are as follows:

[0030] Sa31, the main controller collects and records the DC signal output by the current acquisition array;

[0031] Sa32, the main controller comprehensively judges whether the DC signal is a fault signal based on the DC signal of the motor during normal operation;

[0032] If it is a fault signal, the corresponding flag is set to 1; if it is a non-fault signal, the corresponding flag is set to 0; one current collector corresponds to one DC signal.

[0033] Sa33: The main controller arranges the fault tags of the DC signal according to the current collector array encoding order and the tags obtained in step Sa32, and generates a fault binary code.

[0034] Sa34 is used to look up the corresponding fault motor code range by performing a mapping table lookup on the fault binary code.

[0035] Furthermore, the KNN agent re-examination method includes the following steps:

[0036] Sb1: Collect historical operating data, maintenance records and fault test data of the digital twin workshop to construct a sample set. Divide the sample set into a training set and a validation set according to the proportion. Train and test the constructed KNN agent to obtain the best KNN agent. Deploy the best KNN agent on the digital twin workshop system.

[0037] Sb2, based on the range of faulty motors received from the main controller at the digital twin workshop system end, obtains the operating data of the faulty motors from the digital twin workshop equipment end to obtain a fault data matrix. ;

[0038] Sb3 analyzes the fault data matrix using a KNN agent. The motor fault probability table is obtained and used as the fault detection result.

[0039] Sb4, the digital twin workshop system sends out fault judgment information and adds the maintenance results to the KNN agent to update the sample set.

[0040] Furthermore, the specific steps to obtain the optimal KNN agent are as follows:

[0041] Sb11, by collecting historical operational data and maintenance records from the digital twin workshop, obtains fault diagnosis samples and normal operation data samples, constructing elements as sample points. The sample set;

[0042] Where vector , Let C represent the operating data of the D-th motor at time i, where C is the number of data samples.

[0043] The operating data includes current, voltage, torque, or speed, and can be a single data type or a combination of multiple data types.

[0044] Sb12, fault experiment data is obtained through fault experiments and integrated into the sample set;

[0045] Sb13, the sample set is divided into four equal parts, and three of them are taken as the training set for training sample points T, and the remaining part is taken as the validation set for validation data points V.

[0046] Perform the following operation: Based on the training set and validation set, calculate the Euclidean distance between each validation data point V and the training data point T. Starting with a small number S, gradually increase the value of K until the error rate shows a significant increasing trend; then, based on the Euclidean distance between V and T... Take K training data points T centered at V and corresponding to the smallest Euclidean distance, count the occurrence frequency and fault information of the fault motor number corresponding to the training data points, and obtain the detection result corresponding to each K value; based on the occurrence frequency and fault information of the fault motor number corresponding to the validation set, calculate the error rate of the detection result under each K value;

[0047] Based on different validation sets, repeat the above operation four times in total;

[0048] The error rate of each K value obtained from each operation is averaged. When the average error rate is minimized, a suitable value for K is obtained.

[0049] Obtain the best KNN agent.

[0050] Furthermore, the specific steps of the KNN agent prediction method are as follows:

[0051] Sc1: When the digital twin workshop system does not receive the initial fault detection result, the system obtains the operating data of each motor over a period of time from the digital twin workshop equipment in the order of motor coding, performs fault prediction, and obtains the fault prediction result.

[0052] Sc2 sends information about motors with a probability of failure in the fault prediction results to maintenance personnel for repair.

[0053] Sc3: For motors that have been repaired, the repair personnel enter the repair results into the digital twin workshop system as the actual fault situation and fault information.

[0054] Sc4 adds the fault prediction results, the actual fault conditions of the corresponding motor, and fault information as new sample points to the sample set, thus updating the sample set.

[0055] Furthermore, the specific steps to obtain the fault prediction results are as follows:

[0056] Sc12, constructing the fault prediction matrix Z , where vector , Here, C represents the data for the D-th motor at time i, and C is the number of samples.

[0057] Sc13: Select sample points Y from the KNN agent that correspond to the motor to which the current fault prediction matrix Z belongs, including sample points with faults and sample points that are operating normally.

[0058] Sc14, calculate the filtered sample points Euclidean distance between the fault prediction matrix Z and the fault prediction matrix Z ;

[0059] Sc15 selects the K sample points centered on the fault prediction matrix Z with the smallest Euclidean distance. ;

[0060] Sc16, the statistically obtained sample points The number of times the motor failed and the motor operated normally was obtained, and the probability of the motor failing was calculated as the failure prediction result.

[0061] Compared with the prior art, the significant advantages of this invention are as follows:

[0062] 1. Significantly reduced hardware costs: This invention adopts an array-style cross-deployment strategy, which realizes centralized monitoring of multiple motors by rationally arranging current transformers and current acquisition devices. Compared with the traditional method of deploying one monitoring device for each device, the number of hardware required by this invention is reduced by nearly half. While ensuring the coverage of the detection range, it effectively reduces the equipment procurement and maintenance costs, and is especially suitable for large-scale motor deployment scenarios.

[0063] 2. Improved Fault Detection Efficiency and Real-Time Performance: This invention employs an array-based hardware initial detection method to collect multiple current signals in parallel. The main controller performs real-time comparison and analysis to quickly generate a fault location binary code, which is then mapped to a specific motor coding range. Compared to traditional solutions, this invention requires less computing power and time, achieving automation and speed from data acquisition to initial location, significantly shortening the time for fault detection and preliminary location.

[0064] 3. Intelligent re-inspection accuracy improves with data accumulation: This invention employs the KNN intelligent re-inspection method in the digital twin workshop system. A sample library is constructed using historical operational data and fault experiment data, and cross-validation is used to optimize the intelligent re-inspection model parameters. As the system's operating time increases and the sample library expands, the diagnostic accuracy of the intelligent re-inspection model gradually improves, demonstrating continuous learning and adaptive optimization capabilities.

[0065] 4. Easy to integrate and deploy for large-scale motors: The hardware modules of this invention use current transformers, current acquisition devices and controllers, and support multiple communication methods. When facing complex workshop network environments, the communication methods can be flexibly adjusted for adaptation. With the adoption of fault binary coding and mapping table lookup scheme, this invention can take into account the actual location of the motor during actual deployment, making it easy to integrate and deploy.

[0066] 5. Enhanced Fault Predictability: This invention employs a dual-layer collaborative mechanism of hardware initial inspection and software re-inspection. This not only enables rapid location and diagnosis of existing faults but also effectively enhances the ability to predict potential faults. During periods when no obvious faults are detected during the array-based hardware initial inspection, the digital twin workshop system utilizes idle computing power to continuously monitor and analyze motor operating data based on KNN agents. Combined with fault precursor data from a historical sample library, it identifies potential early abnormal trends in the motor, thereby generating fault prediction information. This predictive capability is continuously optimized as the sample library expands, enabling early warning of potential fault risks. This assists maintenance personnel in taking preventative maintenance measures, effectively avoiding unplanned downtime and improving the continuity and safety of production in the digital twin workshop. Attached Figure Description

[0067] Figure 1 This is a flowchart of the AC motor fault hardware and software collaborative detection method of the present invention;

[0068] Figure 2 This is a schematic diagram of the hardware architecture of the array-type hardware initial inspection method of the present invention;

[0069] Figure 3 This is a flowchart of the KNN intelligent agent re-examination method of the present invention;

[0070] Figure 4 This is a schematic diagram of the array-based cross-deployment strategy of the present invention;

[0071] Figure 5 A graph showing the average error rate for the K-value;

[0072] Figure 6 This is a schematic diagram of the KNN agent re-examination process. Detailed Implementation

[0073] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0074] This invention can be implemented in many different forms and should not be considered as limited to the embodiments described herein. Rather, these embodiments are provided so that the invention will be thoroughly and completely disclosed and will fully express the scope of the invention to those skilled in the art.

[0075] This invention provides a hardware and software collaborative detection method for AC motor faults in a digital twin workshop, such as... Figure 1 As shown, the steps are as follows:

[0076] Step 1: Use an array-based hardware initial inspection method for AC motor faults in a digital twin workshop to obtain the initial fault location results of the AC motor; send the initial fault location results to the twin workshop system via the main controller.

[0077] like Figure 2 As shown, the implementation steps of the array-based hardware initial inspection method are as follows:

[0078] Step a1: Collect information on all motors requiring fault detection at the equipment end of the digital twin workshop, and assign serial numbers to the motors to be inspected to obtain a motor coding table for all motors to be inspected; the specific operation steps are as follows:

[0079] Step a11: Collect information on all motors to be inspected in the digital twin workshop, including: motor type, motor model, motor current and voltage under normal operating conditions, motor physical location, motor drive network, etc., to form a complete motor information database;

[0080] Step a12: Encode all motors to be inspected using the basic serial number within the category to obtain a motor coding table containing all motors to be inspected.

[0081] Step a2: To reduce the number of sensors used, this invention employs an array-based crossover strategy and deploys a current acquisition array based on the motor coding table obtained in step a1 to collect data on motor operation within the workshop; the specific operation steps are as follows:

[0082] Step a21: Based on the motor coding table, obtain the number of motors, the number of current collectors and current transformers required for deploying the current collector array, where one current collector corresponds to one current transformer; the formula for calculating the number of current collectors is as follows:

[0083] ,

[0084] In the formula, The number of current collectors required to deploy a current collector array for type A motors. The number of motors of type A is the same as the number of current transformers and current collectors.

[0085] Step a22: Determine the current acquisition array deployment strategy according to the array-based cross-deployment strategy, as follows:

[0086] Let the motor code in the motor coding table be MOTOR(n), where n has a minimum value of 1 and a maximum value of 1. consecutive positive integers;

[0087] right Each current collector is sequentially encoded, and the encoding of the current collector is represented by m, where m has a minimum value of 1 and a maximum value of 0. A series of positive integers are used to obtain the current collector array;

[0088] The specific relationship between the current acquisition unit and the motor being acquired in the array-style cross-deployment strategy is as follows:

[0089] Current collector ( Collect the total current of all motors;

[0090] When m is not At that time, the m-th current collector collects the motor MOTOR(m) and motor MOTOR(m). )

[0091] And the total current of all motors whose codes are between m and (m+N-1).

[0092] Step a23: Measure the magnitude of the current collected by the current collector array; taking into account the actual operating current of the motor, select a suitable current transformer to convert the motor current into a current that can be collected by the current collector.

[0093] Step a24: The current acquisition device acquires the current after conversion by the current transformer and converts the AC current signal of the motor into a DC signal.

[0094] Step a3: The main controller acquires the DC signal obtained in step a2, analyzes the DC signal, performs initial fault location, and obtains the faulty motor coding range; the specific operation steps are as follows:

[0095] Step a31: The main controller acquires and records the DC signal output by the current acquisition array;

[0096] Step a32: The main controller comprehensively judges whether the DC signal is a fault signal based on the DC signal of the motor operating normally.

[0097] If it is a fault signal, the corresponding flag is set to 1; if it is a non-fault signal, the corresponding flag is set to 0. One current collector corresponds to one DC signal.

[0098] Step a33: The main controller arranges the fault tags of the DC signal according to the current acquisition array encoding order and the tags obtained in step a32, and generates a fault binary code.

[0099] Step a34: Perform a lookup in the mapping table on the fault binary code obtained in step a33 to obtain the corresponding fault motor code range.

[0100] Step a4: Send the faulty motor coding range to the digital twin workshop system as the initial fault detection result.

[0101] Step 2: When the digital twin workshop system receives the initial fault detection result, it uses the KNN intelligent agent re-inspection method to obtain the fault detection result; based on the fault detection result, the digital twin workshop system generates fault judgment information.

[0102] like Figure 3 As shown, the KNN agent re-examination method includes the following steps:

[0103] Step b1: Collect historical operating data, maintenance records, and fault experiment data from the digital twin workshop to construct a sample set, which is then divided into a training set and a validation set according to a set ratio. Deploy the constructed KNN (K-Nearest Neighbor) agent on the digital twin workshop system. Specific operation steps are as follows:

[0104] Step b11 involves collecting historical operational data and maintenance records from the digital twin workshop to obtain fault diagnosis samples and normal operation data samples, constructing elements as sample points. The sample set;

[0105] Where vector , Let C represent the operating data of the D-th motor at time i, where C is the number of data samples.

[0106] The operating data can be any type of data that can be used to diagnose motor faults, such as current, voltage, torque, speed, etc.; the operating data is not limited to a single type of data and can be a combination of multiple data types.

[0107] Historical operating data refers to the data generated by the motor of industrial equipment during actual operation. When the motor shows signs of failure each time, the operator saves the historical operating data for a period of time before the occurrence of the fault signs, and obtains a fault diagnosis sample by marking the corresponding fault type label.

[0108] Step b12: In actual production operations, the faults generated by motors in industrial equipment are unevenly distributed in time and space, and have the characteristics of many types of faults and unpredictable frequency of occurrence. As a result, for fault types with a large time span, it is impossible to obtain sufficient sample data in historical operation. Therefore, fault experiments are conducted to obtain sufficient sample data.

[0109] Fault experiment data will be obtained through fault experiments and integrated into the sample set;

[0110] Among them, fault test data refers to the data generated when the equipment is operated by artificially causing a predetermined fault.

[0111] Step b13: To obtain the value of parameter K in the KNN agent, this invention performs cross-validation on the K value:

[0112] Divide the sample set into four equal parts, take three of them as the training set with elements as training sample points T, and take the remaining part as the validation set with elements as validation data points V.

[0113] Based on the training and validation sets, calculate the Euclidean distance between each validation data point V and the training data point T. The process is represented as follows:

[0114] Calculate vectors and The Euclidean distance between them is:

[0115] ,

[0116] Depend on Obtain the matrix and The Euclidean distance between them is:

[0117] ,

[0118] Where vector , Let C be the operating data of the D-th motor at time i, and C be the number of data samples; vector , The operating data of the H-th motor at time i.

[0119] Starting with a small number S, gradually increase the value of K until K is large enough that the error rate shows a significant increasing trend; then, based on the Euclidean distance between V and T... Take K training data points T with V as the center and the minimum Euclidean distance, count the occurrence frequency and fault information of the fault motor number corresponding to the training data points, and obtain the detection result corresponding to each K value;

[0120] Based on the occurrence frequency and fault information of the fault motor number corresponding to the validation set, the error rate of the detection result under each K value is calculated;

[0121] The above operation is performed four times in total, selecting different parts of the sample set based on the validation set.

[0122] The error rate of each K value obtained from each operation is averaged. When the average error rate is minimized, a suitable value for K is obtained.

[0123] A KNN agent suitable for the digital twin workshop is obtained;

[0124] Step b14: Deploy the KNN agent on the digital twin workshop system.

[0125] Step b2: Based on the range of faulty motors received from the main controller at the digital twin workshop system end, obtain the operating data of the faulty motors from the digital twin workshop equipment end to obtain the fault data matrix. The specific operating steps are as follows:

[0126] Step b21: The digital twin workshop system receives and records the range of faulty motors sent by a main controller at regular intervals.

[0127] Step b22: The digital twin workshop system obtains the operating data of the motors within the range of the faulty motor from the digital twin workshop equipment over a period of time, and constructs a fault data matrix. , where vector , Here, C represents the data for the D-th motor at time i, and C is the number of samples.

[0128] The dimensions, data types, sampling intervals, and sampling times of the running data should be the same as those of the data used to construct the sample set.

[0129] Step b3: Analyze the fault data matrix using a KNN agent. The motor fault probability table is obtained; the specific operation steps are as follows:

[0130] Step b31: Based on the initial fault location results, select sample points in the sample set that belong to the range of the faulty motor. ;

[0131] Step b32, calculate the filtered sample points With fault data matrix Euclidean distance between ;

[0132] Step b33, take The K sample points centered at the center and with the smallest Euclidean distance ;

[0133] Step b34, statistical sample points The frequency of occurrence of the corresponding faulty motor number is calculated to obtain the motor fault probability table, which is the fault detection result.

[0134] Step b4: The digital twin workshop system sends out fault diagnosis information and adds the inspection results to the KNN agent, updating the sample set; the specific operation steps are as follows:

[0135] Step b41: Based on the motor failure probability table and the corresponding failure information, send maintenance suggestions to the maintenance personnel;

[0136] Step b42: The maintenance personnel enter the maintenance results, namely the actual faulty motor and its fault information, into the digital twin workshop system.

[0137] Step b43: Add the fault data matrix and the corresponding actual faulty motor and fault information from this test to the sample set as new sample points, and update the sample set.

[0138] Step 3: When the digital twin workshop system does not receive the initial fault detection result, it uses the KNN agent prediction method to detect fault symptoms and obtain the fault prediction result.

[0139] The specific steps of the KNN agent prediction method are as follows:

[0140] Step c1: If the digital twin workshop system does not receive the initial fault detection result, i.e., there is no fault that has occurred, the digital twin workshop system acquires the operating data of each motor over a period of time in the order of motor codes to perform fault prediction; the specific steps are as follows:

[0141] Step c12, construct the fault prediction matrix Z , where vector , Here, C represents the data for the D-th motor at time i, and C is the number of samples.

[0142] Step c13: Select sample points Y in the KNN agent that correspond to the motor to which the current fault prediction matrix Z belongs, including sample points with faults and sample points that are operating normally.

[0143] Step c14: Calculate the filtered sample points Euclidean distance between the fault prediction matrix Z and the fault prediction matrix Z ;

[0144] Step c15: Select the K sample points centered on the fault prediction matrix Z with the smallest Euclidean distance. ;

[0145] Step c16: Statistically analyze the obtained sample points. The number of times the motor fails and the motor operates normally is obtained, and the probability of the motor failing is calculated, which is the fault prediction result.

[0146] Step c2: Send the information of motors with a probability of failure in the fault prediction results to the maintenance personnel for repair.

[0147] Step c3: For the motors that have been repaired, the repair personnel enter the repair results, namely the actual fault situation and its fault information, into the digital twin workshop system.

[0148] Step c4: Add the fault prediction results, the actual fault conditions of the corresponding motor, and the fault information as new sample points to the sample set, and update the sample set.

[0149] Example 1: Detecting AC motor faults in a digital twin workshop;

[0150] Step E1, motor coding;

[0151] Workshop maintenance personnel compiled digital twin workshop production equipment information. The workshop has eight AC motors of two different performance pairs each. The array-style cross-deployment strategy of this invention was applied to their physical location distribution and wiring methods. Sequential coding was performed on these two types of motors, and the results are shown in Table 1.

[0152] Table 1 Motor Coding Table

[0153]

[0154] Step E2, Hardware connection of an array-based hardware initial inspection method for AC motor faults in a digital twin workshop;

[0155] For a type of coded motor, according to Figure 4 The current collectors are installed using an array-style layout strategy and coded sequentially as A, B, C, D, and E. Each current collector is arranged according to... Figure 4 The array-based detection is shown, specifically as follows:

[0156] The current acquisition unit (A) is responsible for detecting the total current of motors 1-4;

[0157] The current acquisition unit (B) is responsible for detecting the total current of motors 2-5;

[0158] The current acquisition unit (C) is responsible for detecting the total current of motors 3-6;

[0159] The current acquisition unit (D) is responsible for detecting the total current of motors 4-7;

[0160] The current collector (E) is responsible for the current of all motors in the array.

[0161] Using a current transformer to connect the motor and the current acquisition device can ensure the safety of all hardware during the acquisition process.

[0162] The main controller performs real-time sampling and detection on the above 5 current acquisition devices, and analyzes the detected data: for information identified as a fault, the corresponding current acquisition device is marked as 1; for information identified as normal, the corresponding current acquisition device is marked as 0; and the current acquisition device is recorded in the coding order of ABCDE, and the corresponding current acquisition device markings are made into a fault binary code.

[0163] For example, if current collectors (E) and (D) detect an abnormality, but other devices do not, the main controller receives the fault binary code 00011.

[0164] When an AC motor experiences a fault, the collected AC motor fault information shows a significant change, and the main controller can use this characteristic to make judgments on the collected data.

[0165] The main controller performs mapping table analysis on the fault binary code. The mapping table used, namely the fault location information table, is shown in Table 2:

[0166] Table 2 Fault Location Information Table

[0167]

[0168] For example, if the current collector (E), current collector (D), and current collector (A) detect an abnormality, but other devices do not detect an abnormality, then the faulty motor code range corresponding to the faulty binary code 10011 is motor 1, 7, and 8.

[0169] The main controller sends the range of faulty motors to the digital twin workshop system via the workshop's Wi-Fi.

[0170] Step E3, KNN agent re-inspection method for AC motor faults in digital twin workshops;

[0171] Obtaining the sample set: Workshop maintenance personnel retrieved the operation records of the digital twin workshop equipment, corresponding to the motor fault conditions in the maintenance records. They selected the normal operation data and historical fault data for each motor, including operating current and operating voltage. This data was sampled at 0.1ms intervals, 400 times, and the resulting data were used to construct the sample points. The sample set is used to record the corresponding fault type and faulty motor;

[0172] Where vector , Let represent the operating data of the D-th motor at time i, and 400 represent the number of data samples.

[0173] After data collection, 200 sample points were obtained for each motor for fault type A and fault type B, and 40 sample points were obtained for fault type C.

[0174] At this time, due to the short operating history and few failures, it is impossible to obtain enough samples of failure type C from the historical data. Therefore, the workshop maintenance personnel need to conduct failure experiments on 8 motors in the workshop. By artificially interfering with the motors, they can induce the motors to experience a predetermined failure type C in order to obtain relevant motor operating data of failure type C. Compared with software simulation, the data obtained by this method can be closer to the actual failure situation.

[0175] The fault experiment data was sampled at a sampling interval of 0.1ms and 400 times. Sample points Y were constructed and added to the sample set. At this time, each motor and each fault type had 200 sample points labeled with the fault type and the faulty motor.

[0176] KNN agent construction: In order to obtain the value of parameter K in the KNN agent, this invention performs cross-validation on the K values ​​corresponding to the sample set;

[0177] The sample set is divided into 4 equal parts, with 50 sample points for each motor and each fault type in each part. One part is taken as the training sample set with elements as training sample points T. The remaining three parts are used to construct the training dataset with elements as training data points T and the validation dataset with elements as validation data points V, respectively.

[0178] Based on the training and validation datasets, calculate the Euclidean distance between each validation data point V and all elements T in the training dataset. The process is represented as follows:

[0179] Calculate vectors and The Euclidean distance between them is:

[0180] ,

[0181] Depend on Obtain the matrix and The Euclidean distance between them is:

[0182] ,

[0183] Starting with a small number 1, gradually increase the value of K until it is large enough; then, based on the Euclidean distance between V and T... Take K training data points T with V as the center and the minimum Euclidean distance, count the occurrence frequency and fault information of the fault motor number corresponding to the training data points, and obtain the detection result corresponding to each K value;

[0184] Based on the occurrence frequency and fault information of the fault motor number corresponding to the validation dataset, the error rate of the detection results under each K value is calculated;

[0185] The above operation is performed four times in total, taking different parts of the training sample set from the validation dataset.

[0186] The error rate of each K value obtained in each operation is averaged, and the result is as follows: Figure 5 As shown, when the average error rate is minimized, the appropriate value for K is 6.

[0187] A KNN agent is constructed using the training sample set and parameters K.

[0188] Fault Detection and Fault Location: Upon receiving the range of faulty motors from the main controller, the digital twin workshop system immediately retrieves motor operating data before and after the initial fault detection from the corresponding motor equipment terminal, including operating current and operating voltage. This data is sampled at 0.1ms intervals and 400 times to construct a fault data matrix. ;

[0189] Where vector , This represents the two-dimensional data of current and voltage of the nth motor at time i, with 400 representing the number of samples.

[0190] Based on the range of faulty motors, sample points Y within the fault range of the labeled faulty motors are selected from the sample set, and the fault data matrix is ​​calculated. Euclidean distance between the selected sample point Y The process is represented as follows:

[0191] Calculate vectors and The Euclidean distance between them is:

[0192] ,

[0193] Depend on Obtain the matrix and The Euclidean distance between them is:

[0194] ,

[0195] for and Euclidean distance between Take The K=6 sample points centered at the center with the smallest Euclidean distance The frequency of occurrence of faulty motor numbers corresponding to sample points is counted, and the frequency of occurrence of faulty motor numbers is calculated to obtain a motor fault probability table, which is the fault detection result. Figure 6 As shown, when the fault range includes motors 1, 7, and 8, the following applies: Centered on the dashed circle, with the boundary as the boundary, the points inside the boundary (including the points on the boundary line) are the obtained sample points. Different shapes are used to represent the fault motor numbers corresponding to the sample points, and different colors are used to represent the fault conditions of the sample points. It is found that motor 1 has 6 faults, motor 7 has 0 faults, and motor 8 has 0 faults. Therefore, motor 1 is the motor with the highest probability of failure, and the fault type is short circuit fault.

[0196] Send a motor failure probability table to maintenance personnel, and send failure judgment information to maintenance personnel based on the failure information corresponding to the sample points and the corresponding motor failure probability table.

[0197] KNN agent self-optimization:

[0198] After the maintenance is completed, the maintenance personnel input the maintenance results, namely the motor number and fault type of the actual fault, into the digital twin workshop system. Based on the maintenance results, the fault data matrix is ​​then processed. The points are marked and added to the sample set as new sample points Y, which improves the positioning accuracy and the accuracy of maintenance opinions when the next fault occurs.

[0199] Example 2: Predicting AC motor faults in the digital twin workshop described in Example 1;

[0200] Step F1 to enter fault prediction mode;

[0201] Once the latest fault detection result is confirmed and received by the maintenance personnel, if the digital twin workshop system does not receive the initial fault detection result sent by the main controller within a certain period of time, the system will enter the fault prediction mode. If the initial fault detection result is received during this period, the system will immediately exit and switch to KNN agent re-inspection.

[0202] Step F2, Fault Prediction;

[0203] The idle computing power available on the digital twin workshop system side supports KNN model operation at a sampling frequency of 200kHz. The system side retrieves motor operation data from the motor equipment terminal one by one according to the motor number order, including the two dimensions of operating current and operating voltage.

[0204] For the current motor, its operating data is sampled at 0.1ms intervals, with 400 samples taken, to construct a fault prediction matrix Z. ;

[0205] Based on the current motor number, select sample points Y from the sample set that are labeled as the current motor, including sample points composed of data from normal operation.

[0206] Calculate the Euclidean distance between the fault prediction matrix Z and the selected sample points Y. K=6 sample points are selected with the smallest Euclidean distance centered on the fault prediction matrix Z. The number of times the fault condition corresponding to the statistical sample point occurs is counted, the probability of the current motor malfunction is calculated, and the fault prediction result is obtained. If all 6 sample points in the range are representative of normal operation, the prediction result is that the current motor has no fault signs and is operating normally.

[0207] Step F3: Prediction results report and feedback;

[0208] For motors predicted to have a probability of failure, the digital twin workshop system reports to the maintenance personnel. After receiving the report, the maintenance personnel can conduct an inspection. For predictions with corresponding maintenance results, the maintenance personnel input the maintenance results, i.e., the actual motor fault status and fault type, into the digital twin workshop system. Based on the maintenance results, the fault prediction matrix Z is marked and becomes a new sample point Y, which is added to the sample set to improve the positioning accuracy and the accuracy of maintenance opinions when the next fault occurs.

[0209] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A hardware and software collaborative detection method for AC motor faults in a digital twin workshop, characterized in that, The steps include the following: An array-based hardware initial inspection method is used to obtain the initial fault location result of the AC motor; the initial fault location result is sent to the digital twin workshop system via the main controller as the initial fault inspection result; When the digital twin workshop system receives the initial fault detection result, it uses the KNN intelligent agent re-inspection method to obtain the fault detection result; Based on the fault detection results, the digital twin workshop system generates fault judgment information; When the digital twin workshop system does not receive the initial fault detection result, the KNN agent prediction method is used to detect fault symptoms and obtain the fault prediction result. Based on the fault prediction results, the AC motor was inspected and repaired, and the actual fault conditions and fault information were recorded. The specific steps of the array-type hardware initial inspection method include: Sa1: Collect information on all motors that need to be fault-detected in the digital twin workshop equipment, and encode the motors to be inspected by serial number to obtain a motor code table for all motors to be inspected. Sa2, based on the motor coding table and array-type cross-strategy, deploys a current acquisition array to collect AC signals from motors running in the workshop and outputs DC signals. Sa3, the main controller collects DC signals, analyzes the DC signals and performs initial fault location to obtain the fault motor coding range; Sa4 sends the faulty motor coding range to the digital twin workshop system. The specific steps for collecting AC electrical signals from motors running in the workshop are as follows: Sa21, based on the motor coding table, yields the number of motors, the number of current collectors required for deploying the current collector array, and the number of current transformers. One current collector corresponds to one current transformer. The formula for calculating the number of current collectors is as follows: ， In the formula, The number of current collectors required to deploy a current collector array for type A motors. The number of motors of type A; Sa22 determines the current acquisition array deployment strategy based on an array-based cross-deployment approach, as follows: Let the motor code in the motor coding table be MOTOR(n), where n is a consecutive positive integer. ; right The current collectors are sequentially encoded to obtain a current collector array. The encoding of each current collector is represented by m, where m is a consecutive positive integer. ; The specific relationship between the current acquisition device and the motor being acquired in the array-type cross-deployment strategy is as follows: 1) Current acquisition device Collect the total current of all motors; 2) When m is not equal to At that time, the m-th current collector collects the motor MOTOR(m) and the motor MOTOR(m). The total current of all motors between ) Sa23 measures the magnitude of the current collected by the current collector array and converts the AC current signal of the motor into a DC signal.

2. The method for collaborative detection of AC motor faults in a digital twin workshop according to claim 1, characterized in that, The specific steps to obtain the faulty motor coding range are as follows: Sa31, the main controller collects and records the DC signal output by the current acquisition array; Sa32, the main controller comprehensively judges whether the DC signal output by the current acquisition array is a fault signal based on the DC signal of the motor working normally; If it is a fault signal, mark the corresponding current collector as 1; If the signal is not a fault signal, the corresponding current collector is marked as 0; where one current collector corresponds to one DC signal. Sa33: The main controller arranges the fault tags of the DC signal according to the current collector array encoding order and the tags obtained in step Sa32, and generates a fault binary code. Sa34 is used to look up the corresponding fault motor code range by performing a mapping table lookup on the fault binary code.

3. The method for collaborative detection of AC motor faults in a digital twin workshop according to claim 1, characterized in that, The KNN agent re-examination method includes the following steps: Sb1: Collect historical operating data, maintenance records and fault test data of the digital twin workshop to construct a sample set. Divide the sample set into a training set and a validation set according to the proportion. Train and test the constructed KNN agent to obtain the best KNN agent. Deploy the best KNN agent on the digital twin workshop system. Sb2, based on the fault motor coding range received from the main controller at the digital twin workshop system end, obtains the operating data of the fault motor from the digital twin workshop equipment end to obtain the fault data matrix. ; Sb3 analyzes the fault data matrix using a KNN agent. The motor fault probability table is obtained and used as the fault detection result. Sb4, the digital twin workshop system sends out fault judgment information and adds the maintenance results to the KNN agent to update the sample set.

4. The method for collaborative detection of AC motor faults in a digital twin workshop according to claim 3, characterized in that, The specific steps to obtain the optimal KNN agent are as follows: Sb11, by collecting historical operational data and maintenance records from the digital twin workshop, obtains fault diagnosis samples and normal operation data samples, constructing elements as sample points. The sample set; Where vector , Let C represent the operating data of the D-th motor at time i, where C is the number of data samples. The operating data includes current, voltage, torque, or speed, and can be a single data type or a combination of multiple data types. Sb12, fault experiment data is obtained through fault experiments and integrated into the sample set; Sb13, the sample set is divided into four equal parts, three of which are taken as the training set for training data point T, and the remaining part is taken as the validation set for validation data point V. Perform the following operation: Based on the training set and validation set, calculate the Euclidean distance between each validation data point V and the training data point T. Starting with a small number S, gradually increase the value of K until the error rate shows a significant increasing trend; then, based on the Euclidean distance between V and T... Take K training data points T centered at V and corresponding to the smallest Euclidean distance, count the occurrence frequency and fault information of the fault motor number corresponding to the training data points, and obtain the detection result corresponding to each K value; based on the occurrence frequency and fault information of the fault motor number corresponding to the validation set, calculate the error rate of the detection result under each K value; Repeat the above operation four times, depending on the different validation sets. The error rate of each K value obtained from each operation is averaged. When the average error rate is minimized, a suitable value for K is obtained. Obtain the best KNN agent.

5. The method for collaborative hardware and software fault detection of AC motors in a digital twin workshop according to claim 1, characterized in that, The specific steps of the KNN agent prediction method are as follows: Sc1: When the digital twin workshop system does not receive the initial fault detection result, the system obtains the operating data of each motor over a period of time from the digital twin workshop equipment in the order of motor coding, performs fault prediction, and obtains the fault prediction result. Sc2 sends information about motors with a probability of failure in the fault prediction results to maintenance personnel for repair. Sc3: For motors that have been repaired, the repair personnel enter the repair results into the digital twin workshop system as the actual fault situation and fault information. Sc4 adds the fault prediction results, the actual fault conditions of the corresponding motor, and fault information as new sample points to the sample set, thus updating the sample set.

6. The method for collaborative hardware and software fault detection of AC motors in a digital twin workshop according to claim 5, characterized in that, The specific steps to obtain the fault prediction results are as follows: Sc12, constructing the fault prediction matrix Z , where vector , Here, C represents the data for the D-th motor at time i, and C is the number of samples. Sc13: Select sample points Y from the KNN agent that correspond to the motor to which the current fault prediction matrix Z belongs, including sample points with faults and sample points that are operating normally. Sc14, calculate the filtered sample points Euclidean distance between the fault prediction matrix Z and the fault prediction matrix Z ; Sc15 selects the K sample points centered on the fault prediction matrix Z with the smallest Euclidean distance. ; Sc16, the statistically obtained sample points The number of times the motor failed and the motor operated normally was obtained, and the probability of the motor failing was calculated as the failure prediction result.

Images