A System and Method for Detecting Motor Anomalies in a Fully Automated Maglev Track Production Line

Abstract

This invention discloses an anomaly detection system and method for a fully automated magnetic levitation track motor, relating to the field of intelligent motor detection. The system includes: a feature acquisition unit for acquiring time-series data of various operational features of the fully automated magnetic levitation track motor; a feature extraction unit for calculating the weight of each operational feature at each time point and extracting operational features based on the weights to obtain an operational feature set; a labeling unit for labeling the operational feature set to obtain a training set; and a detection unit for constructing a first motor anomaly detection model, training the first motor anomaly detection model to obtain a second motor anomaly detection model, the second motor anomaly detection model being used to output motor detection results. The operational features include the motion characteristics of the float of the fully automated magnetic levitation track motor. This invention extracts an adaptive feature set based on time-series data, which can improve the accuracy of system detection.

Description

Technical Field

[0001] This invention relates to the field of intelligent motor detection, specifically to a system and method for detecting abnormal motors in a fully automated magnetic levitation track production line. Background Technology

[0002] A fully automated magnetic levitation conveyor line is a transport system that uses the principle of magnetic levitation to move goods on a suspended platform, and it is gradually being applied in industrial production. The magnetic levitation planar motor is the core equipment that enables its function. The magnetic levitation planar motor consists of a stationary stator, a float suspended on the stator, and a corresponding control system. Without mechanical guide rail support, it can directly achieve large-stroke two-dimensional planar drive of the float, and is small in size and light in weight, enabling precise movement under vacuum conditions. The operating status of the magnetic levitation planar motor directly affects the accuracy and stability of the conveyor line's material transport. Therefore, it is necessary to monitor the motor's operating status.

[0003] Currently, anomaly detection in fully automated maglev rail production lines still relies heavily on experience-based judgment, monitoring key state parameters by setting thresholds. However, this method has limitations. For example, threshold setting lacks a unified standard, heavily depends on human experience, and often only targets known, obvious fault characteristics, easily missing potential fault signals. Secondly, motors operate in diverse conditions, and statically preset thresholds cannot flexibly handle this complexity, thus affecting detection accuracy. Furthermore, the scarcity of anomaly data and the complex correlations between data points make it difficult to construct accurate and reliable detection models. Summary of the Invention

[0004] The purpose of this invention is to construct a fully automated motor anomaly detection system for magnetic levitation track production lines, thereby improving the reliability and accuracy of motor anomaly detection.

[0005] To achieve the above objectives, the present invention provides a fully automated motor anomaly detection system for a magnetic levitation track assembly line, the system comprising:

[0006] The feature acquisition unit is used to collect time-series data of various operational features during the operation of the motor in the fully automated magnetic levitation track production line.

[0007] The feature extraction unit is used to calculate the weight of each of the operational features at each time point based on the time series data, and extract several operational features based on the weights to obtain the operational feature set for the corresponding time point;

[0008] The annotation unit is used to annotate the running feature set at all time points to obtain the training set;

[0009] The detection unit is used to construct a first motor anomaly detection model, and to train the first motor anomaly detection model based on the training set to obtain a second motor anomaly detection model. The second motor anomaly detection model is used to identify and extract the operating characteristics of the motor to be detected, and output the motor detection result.

[0010] Among them, the motion characteristics of the float of the magnetic levitation planar motor, such as running speed and running acceleration, can also reflect the running state of the motor. Therefore, the running characteristics described in this invention include the motion characteristics of the float of the fully automatic magnetic levitation track assembly line motor.

[0011] The principle of this invention is as follows: The operating conditions of the motors in a fully automated magnetic levitation track production line are complex and variable, with inherent complexity and diversity in their operating characteristics. Furthermore, as operating time increases, the motor's performance gradually declines, leading to further changes in its operating characteristics. Traditional motor detection model training often relies on a fixed feature set, making it difficult for the model to cope with these changes and resulting in incorrect detection results. A fixed feature set also prevents the model from effectively learning the underlying patterns and features of the data, limiting its generalization ability. Therefore, in the system of this invention, the feature extraction unit determines the weight of the operating feature at each time point based on the time-series data of the motor's operating characteristics, and extracts the key features that best reflect the motor's operating state according to the weight, constructing an operating feature set. This process achieves adaptive dynamic adjustment of the feature set, reducing noise and redundant features, and allowing the training set to contain richer and more diverse training data. Consequently, the model can more accurately capture the potential changes in operating characteristics as the motor's operating time increases, giving the model stronger generalization ability and improving the accuracy and robustness of motor detection.

[0012] Furthermore, the method for calculating the weight of each operational feature at each time point based on the time series data is as follows:

[0013]

[0014] Where, ω i (t) and ω i (t-1) represent the weights of the i-th operational feature at time point t and time point t-1, respectively, α is the balance factor, and d i For the time series data of the i-th running feature before time point t, var(d i ) and mean(d i ) are respectively d i The variance and mean of . The larger the value, the more dispersed the time series data of the corresponding operating feature is. This indicates that the operating feature varies greatly during motor operation, meaning it is more sensitive to changes in the motor's operating state and therefore receives a larger weight. The balance factor is used to control the stability of weight changes. When the balance factor is large, the weight of the previous time point has a greater impact on the weight of the current time point; conversely, the dispersion of the time series data of the operating feature has a greater impact on the weight of the current time point.

[0015] Furthermore, the step of extracting several operational features based on the weights to obtain an operational feature set for the corresponding time point includes:

[0016] The operational features at the corresponding time points are sorted in descending order of weight.

[0017] The top n operational features are selected as the adaptive feature set F(t) at time point t;

[0018] The running features in the adaptive feature set F(t) are standardized respectively to obtain the standardized values ​​of the running features;

[0019] The standardized value is used as the running feature set F(t) at time point t.

[0020] Among these, the top n operational features best reflect the motor's operating state and are therefore used to construct the adaptive feature set. Since the adaptive feature set eliminates operational features with excessively low weights, redundant information is reduced, lowering the computational complexity of model training and improving training efficiency. Because the motor's operational features involve multiple different physical quantities, standardization can transform these features to the same dimension and scale, which is beneficial for subsequent analysis and processing.

[0021] Furthermore, the running features in the adaptive feature set F(t) are standardized using the following formula:

[0022]

[0023] Among them, f i For the original data of the i-th running feature, f i , for f i The standardized value, std(d i ) is d i The standard deviation.

[0024] Furthermore, the second motor anomaly detection model is an LSTM model, and the motor detection results output by the second motor anomaly detection model include: motor operating state A.

[0025] The LSTM model, by introducing gating mechanisms such as input gates, forget gates, and output gates, and maintaining a cell state, effectively controls the flow of information, thus mitigating the vanishing and exploding gradient problems. This allows LSTM to capture long-range dependencies in long-term series and can be applied to various time series tasks. Time series data is typically non-linear and contains noise and other non-linear features. The LSTM model has the ability to learn non-linear functions, enabling it to better capture these non-linear features and achieve more accurate predictions.

[0026] Furthermore, the system also includes an update unit, which is used to update the model parameters of the second motor anomaly detection model. The update method for the model parameters is as follows:

[0027]

[0028] Where Θ(t) and Θ(t-1) are the model parameters of the second motor anomaly detection model at time point t and time point t-1, respectively, β(t) is the learning rate at time point t, and J is the loss function. By updating the model parameters, the model can better fit the training data, thereby improving the accuracy of the detection results; updating the model parameters can also make the model perform better on unseen data, enhancing the model's generalization ability. This invention can also retain a portion of the training set during the training of the second motor anomaly detection model to train the updated model, thereby enhancing the model's long-term memory capability.

[0029] Furthermore, β(t) is calculated as follows:

[0030]

[0031] Where β(t-1) is the learning rate at time point t-1, γ is the learning rate adjustment factor, and A(t) and A(t-1) are the operating states of the motor at time points t and t-1, respectively, output by the second motor anomaly detection model. This invention continuously optimizes the model parameters by dynamically adjusting the learning rate, thereby enabling the model to better adapt to real-time changes in the motor and improving the system's applicability and robustness.

[0032] Furthermore, the loss function J is calculated as follows:

[0033] J(t) = |A(t) - S(t)|;

[0034] Where S(t) represents the actual state of the motor at time t.

[0035] This invention also provides a method for detecting motor malfunctions in a fully automated magnetic levitation track production line, comprising the following steps:

[0036] Time-series data of various operational characteristics of the motors in the fully automated production line of the magnetic levitation track were collected.

[0037] The weights of the operational features at each time point are calculated based on the time series data, and several operational features are extracted based on the weights to obtain the operational feature set for the corresponding time point.

[0038] The runtime feature sets at all time points are labeled to obtain the first training set and the second training set;

[0039] A first motor anomaly detection model is constructed, and a second motor anomaly detection model is obtained by training the first motor anomaly detection model based on the first training set. The second motor anomaly detection model is used to identify and extract the operating characteristics of the motor to be detected and output the motor detection results.

[0040] Furthermore, the method also includes the following steps:

[0041] Update the model parameters of the second motor anomaly detection model;

[0042] The second motor anomaly detection model with updated model parameters is retrained based on the second training set.

[0043] One or more technical solutions provided by this invention have at least the following technical effects or advantages:

[0044] 1. This invention extracts an adaptive feature set based on time series data to obtain an adaptive feature set that effectively reflects the motor's operating state, which can improve the accuracy of system detection. By dynamically adjusting the feature set, the model can adapt to the changes in operating characteristics as the motor's operating time increases, thereby enhancing the model's generalization ability.

[0045] 2. In the process of updating model parameters, this invention implements a strategy of dynamically adjusting the learning rate to achieve more refined model parameter optimization. Attached Figure Description

[0046] The accompanying drawings, which are provided to further illustrate embodiments of the invention and constitute a part of this invention, are not intended to limit the scope of the invention.

[0047] Figure 1 This is a schematic diagram of the composition of a fully automated magnetic levitation track motor anomaly detection system according to the present invention;

[0048] Figure 2 This is a flowchart illustrating a detection method for an abnormal motor detection system based on a fully automated production line using magnetic levitation tracks, as described in this invention. Detailed Implementation

[0049] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, where there is no conflict, the embodiments of the present invention and the features thereof can be combined with each other.

[0050] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0051] Example 1

[0052] Please refer to Figure 1 Embodiment 1 of the present invention provides a fully automated motor anomaly detection system for a magnetic levitation track assembly line, the system comprising:

[0053] The feature acquisition unit is used to collect time-series data of various operational features during the operation of the motor in the fully automated magnetic levitation track production line.

[0054] The feature extraction unit is used to calculate the weight of each of the operational features at each time point based on the time series data, and extract several operational features based on the weights to obtain the operational feature set for the corresponding time point;

[0055] The annotation unit is used to annotate the running feature set at all time points to obtain the training set;

[0056] The detection unit is used to construct a first motor anomaly detection model, and to train the first motor anomaly detection model based on the training set to obtain a second motor anomaly detection model. The second motor anomaly detection model is used to identify and extract the operating characteristics of the motor to be detected, and output the motor detection result.

[0057] The magnetic levitation planar motor primarily relies on electromagnetic force to levitate and move the float. Specifically, the motor's control system supplies current to the stator coils, generating a magnetic field. This magnetic field interacts with the permanent magnets in the float, producing electromagnetic force. By altering the characteristics of the current, the float can levitate and move without mechanical contact. The current characteristics, along with the float's load and motion characteristics, reflect the operating state of the magnetic levitation planar motor.

[0058] Therefore, in this embodiment, the operating features collected by the feature acquisition unit include the motion features of the float of the fully automatic assembly line motor of the magnetic levitation track, the load of the float and the current features flowing through the coil. The motion features of the float include the running speed and acceleration of the float, and the current features include the maximum value, minimum value, median, mean, variance, peak value, peak-to-peak value, effective value, peak factor, margin factor, impulse factor, waveform factor, kurtosis factor and skewness factor.

[0059] The buoy's speed is measured by a speed sensor, preferably a laser-type speed sensor; the buoy's acceleration is acquired by an accelerometer; the current characteristics are obtained by a current sensor, which collects electrical signals and then analyzes and calculates them using computer software. Preferably, the current sensor is a Hall effect sensor or an open-loop current sensor, and the computer software is preferably MATLAB. Those skilled in the art can select the specific model of the aforementioned sensors as needed, and this invention does not limit the choice.

[0060] It should be noted that in actual fully automated magnetic levitation rail production lines, in some cases, each float only undertakes a partial independent transport task, such as transferring a specific part to the next processing area. After processing, the part is then transferred to the next area by another float. In this case, the running paths and motion characteristics of different floats on the same motor are different. If the complete motion characteristics of each float are collected, the computational complexity will be greatly increased, and the cost of model training will be increased.

[0061] Therefore, this embodiment provides a method for extracting motion features to solve the above problems. Specifically, it includes the following steps:

[0062] The processing cycle of the part to be processed in the fully automated production line of the magnetic levitation track is obtained, that is, the total time taken for the part to be processed from entering the starting station to leaving the end station.

[0063] For different floats, the processing cycle is divided into different number of time periods. For example, for float A, it is divided into 9 time periods, and for float B, it is divided into 12 time periods, so that in each time period, the motion characteristics (such as speed) of the corresponding float have at most one maximum value or one minimum value.

[0064] The maximum, minimum and principal slope of the motion characteristics of each float are obtained in each time period, wherein the principal slope is the maximum value minus the minimum value divided by the length of the time period;

[0065] The maximum value, the minimum value, and the principal slope are used as feature vectors of the corresponding motion characteristics of the float.

[0066] During continuous operation of the production line, the movement of each float exhibits a periodic characteristic consistent with the processing cycle of the parts to be processed. Therefore, this invention uses the processing cycle as the basis for dividing time periods. Based on the movement of different floats, the processing cycle is divided into different numbers of time periods. For example, floats with complex movement patterns exhibit large fluctuations in their movement characteristics, thus requiring more time periods. This reduces the number of data points, lowers computational complexity, and ensures that the maximum, minimum, and principal slopes effectively reflect the trends and changes in the corresponding movement characteristics, thereby maintaining detection accuracy.

[0067] The calculation method for the weight of each operational feature at each time point based on the time series data is as follows:

[0068]

[0069] Where, ω i (t) and ω i (t-1) represent the weights of the i-th operational feature at time point t and time point t-1, respectively, α is the balance factor, and d i For the time series data of the i-th running feature before time point t, var(d i ) and mean(d i ) are respectively d i The variance and mean of .

[0070] The initial weights of the operating features and the values ​​of the balance factor α can be set by those skilled in the art based on experience and actual conditions. Through the above formula, this invention effectively improves upon the limitations of static weights, dynamically adjusting the weights of each part according to the real-time changes in the operating features, enabling the model to promptly capture new trends in the data and improving the model's adaptability and predictive accuracy.

[0071] The step of extracting several operational features based on the weights to obtain an operational feature set for a corresponding time point includes:

[0072] The operational features at the corresponding time points are sorted in descending order of weight.

[0073] The top n operational features are selected as the adaptive feature set F(t) at time point t;

[0074] The running features in the adaptive feature set F(t) are standardized respectively to obtain the standardized values ​​of the running features;

[0075] The standardized value is used as the running feature set F(t) at time point t.

[0076] The specific value of n can be set by those skilled in the art based on the actual situation.

[0077] In this embodiment, Z-score standardization is used to standardize the running features in the adaptive feature set F(t). The specific formula is as follows:

[0078]

[0079] Among them, f i For the original data of the i-th running feature, f i , for f i The standardized value, std(d i ) is d i The standard deviation.

[0080] The annotation unit labels the running feature set F(t) of all time points as either normal or abnormal motor operation based on the corresponding motor status.

[0081] In this embodiment, the second motor anomaly detection model is an LSTM model, and the motor detection results output by the second motor anomaly detection model include: motor operating state A.

[0082] The system further includes an update unit, which is used to update the model parameters of the second motor anomaly detection model. The update method for the model parameters is as follows:

[0083]

[0084] Where Θ(t) and Θ(t-1) are the model parameters of the second motor anomaly detection model at time point t and time point t-1, respectively, β(t) is the learning rate at time point t, and J is the loss function. In this embodiment, the model parameters Θ include the input feature dimension, the number of hidden layer features, the number of LSTM stacked layers, bias, batch processing settings, Dropout layer, bidirectional mode, etc., and the initial parameters can be set according to the actual situation.

[0085] The calculation method for β(t) is as follows:

[0086]

[0087] Where β(t-1) is the learning rate at time point t-1, and γ is the learning rate adjustment factor, the specific values ​​of which can be determined by those skilled in the art based on experience and actual conditions; A(t) and A(t-1) are the operating states of the motor at time points t and t-1, respectively, output by the second motor anomaly detection model (t≥1). This invention, by dynamically adjusting the learning rate, enables the LSTM model to adapt more flexibly to changes in data, improves the model's sensitivity to real-time changes in motor operating characteristics or performance, and enhances the system's applicability and robustness.

[0088] The loss function J is calculated as follows:

[0089] J(t) = |A(t) - S(t)|;

[0090] Where S(t) represents the actual state of the motor at time t.

[0091] Example 2

[0092] Please refer to Figure 2 This embodiment provides a detection method based on the magnetic levitation track fully automated assembly line motor anomaly detection system in Embodiment 1, including the following steps:

[0093] Time-series data of various operational characteristics of the motors in the fully automated production line of the magnetic levitation track were collected.

[0094] The weights of the operational features at each time point are calculated based on the time series data, and several operational features are extracted based on the weights to obtain the operational feature set for the corresponding time point.

[0095] The runtime feature sets at all time points are labeled to obtain the first training set and the second training set;

[0096] A first motor anomaly detection model is constructed, and a second motor anomaly detection model is obtained by training the first motor anomaly detection model based on the first training set. The second motor anomaly detection model is used to identify and extract the operating characteristics of the motor to be detected and output the motor detection results.

[0097] The method further includes the following steps:

[0098] Update the model parameters of the second motor anomaly detection model;

[0099] The second motor anomaly detection model with updated model parameters is retrained based on the second training set.

[0100] In this embodiment, the training set is divided into a first training set and a second training set. The first training set is used to train the anomaly detection model for the first motor, while the second training set is retained to train the updated anomaly detection model for the second motor, so as to improve the long-term memory capability of the second motor anomaly detection model.

[0101] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0102] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A fully automated magnetic levitation track assembly line motor anomaly detection system, characterized in that, The system includes: The feature acquisition unit is used to collect time-series data of various operational features during the operation of the motor in the fully automated magnetic levitation track production line. The feature extraction unit is used to calculate the weight of each of the operational features at each time point based on the time series data, and extract several operational features based on the weights to obtain the operational feature set for the corresponding time point; The annotation unit is used to annotate the running feature set at all time points to obtain the training set; The detection unit is used to construct a first motor anomaly detection model, and to train the first motor anomaly detection model based on the training set to obtain a second motor anomaly detection model. The second motor anomaly detection model is used to identify and extract the operating characteristics of the motor to be detected, and output the motor detection result. The operational characteristics include the motion characteristics of the float of the fully automated magnetic levitation track assembly line motor; The method for calculating the weight of each operational feature at each time point based on the time series data is as follows: ； in, and The i-th operational feature is located at time point. and time point The weight, As a balance factor, For the i-th operational feature at time point Previous time series data, and They are respectively The variance and mean of .

2. The magnetic levitation track fully automated assembly line motor anomaly detection system according to claim 1, characterized in that, The step of extracting several operational features based on the weights to obtain an operational feature set for a corresponding time point includes: The operational features at the corresponding time points are sorted in descending order of weight. The top n operational features were selected as time points. Adaptive feature set ; For the adaptive feature set respectively The operational characteristics described in the data are standardized to obtain standardized values ​​for the operational characteristics. Using the standardized value as the time point The set of operational features .

3. The magnetic levitation track fully automated assembly line motor anomaly detection system according to claim 2, characterized in that, The adaptive feature set is applied using the following formula. The operational characteristics mentioned above are standardized: ； in, The original data for the i-th operational feature. for The standardized value, for The standard deviation.

4. The magnetic levitation track fully automated assembly line motor anomaly detection system according to claim 1, characterized in that, The second motor anomaly detection model is an LSTM model. The motor detection results output by the second motor anomaly detection model include: motor operating status. .

5. The magnetic levitation track fully automated assembly line motor anomaly detection system according to claim 4, characterized in that, The system further includes an update unit, which is used to update the model parameters of the second motor anomaly detection model. The update method for the model parameters is as follows: ； in, and The second motor anomaly detection model is located at time points. and time point Model parameters, For time points The learning rate This is the loss function.

6. The abnormality detection system for a fully automated magnetic levitation track motor as described in claim 5, characterized in that, The calculation method is as follows: ； in, For time points The learning rate This is the learning rate adjustment factor. and The motor at time points output by the second motor anomaly detection model are respectively... and time point The running status.

7. The abnormality detection system for a fully automated magnetic levitation track motor as described in claim 5, characterized in that, loss function The calculation method is as follows: ； in, For the motor at a certain time point The true state.

8. A method for detecting motor malfunctions in a fully automated magnetic levitation track production line, characterized in that, Includes the following steps: Time-series data of various operational characteristics of the motors in the fully automated production line of the magnetic levitation track were collected. The weights of the operational features at each time point are calculated based on the time series data, and several operational features are extracted based on the weights to obtain the operational feature set for the corresponding time point. The runtime feature sets at all time points are labeled to obtain the first training set and the second training set; A first motor anomaly detection model is constructed, and a second motor anomaly detection model is obtained by training the first motor anomaly detection model based on the first training set. The second motor anomaly detection model is used to identify and extract the operating characteristics of the motor to be detected and output the motor detection result. The method for calculating the weight of each operational feature at each time point based on the time series data is as follows: ； in, and The i-th operational feature is located at time point. and time point The weight, As a balance factor, For the i-th operational feature at time point Previous time series data, and They are respectively The variance and mean of .

9. The method for detecting motor malfunctions in a fully automated magnetic levitation track assembly line according to claim 8, characterized in that, The method further includes the following steps: Update the model parameters of the second motor anomaly detection model; The second motor anomaly detection model with updated model parameters is retrained based on the second training set.

Images