Operation data analysis method based on motor control

Abstract

The invention discloses a motor control-based operation data analysis method, which comprises the following steps of: acquiring current motor sensing streaming information, segmenting the current motor sensing streaming information, and performing representation information mining on sensing streaming information segments to obtain an operation state description vector of the current motor sensing streaming information; identifying a to-be-determined motor operation state sequence corresponding to the current motor sensing streaming information, determining and obtaining at least one state significant data point in the current motor sensing streaming information, determining at least one positioning interval of state indication information flow in the current motor sensing streaming information based on the state significant data point, and determining the to-be-determined motor operation state sequence according to the positioning interval. The operation state description vectors are extracted according to the positioning interval, the target operation state description vector of each state indication information flow is obtained, the to-be-determined motor operation state sequence is optimized based on the target operation state description vectors, and the final motor operation state sequence is obtained. And accurate motor control is facilitated.

Description

Technical Field

[0001] The present invention belongs to the technical field of motor operation data processing, and particularly relates to a method for analyzing operation data based on motor control. Background Art

[0002] In the industrial production system, as a key power output device, the stable and efficient operation of the motor plays a decisive role in the smooth progress of the entire production process. Accurately monitoring the motor operation status and implementing effective control can significantly reduce equipment failure rates, improve production efficiency, and reduce energy consumption and maintenance costs.

[0003] The existing motor operation status monitoring and control technologies have many limitations: 1) Single data dimension: Some monitoring methods only rely on a single type of sensing data, such as analyzing only based on current data or voltage data. The operation of the motor is a complex process, and single data cannot comprehensively reflect the actual operation status of the motor. For example, in the initial stage of a motor bearing failure, the current and voltage may not change significantly, but the temperature and speed have already shown abnormal fluctuations. Relying only on electrical parameter monitoring will miss such failures.

[0004] 2) Weak ability to handle sudden states: Traditional technologies have difficulty accurately grasping the evolution process of the motor operation status when dealing with sudden states of the motor. When the motor suddenly overloads or has abnormal speed, traditional methods cannot timely and accurately judge the starting, developing, and ending stages of the failure, resulting in a lag in control decisions and being unable to effectively avoid equipment damage.

[0005] 3) Insufficient data feature mining: When processing sensing data, there is a lack of in-depth mining of data features, resulting in low accuracy and reliability of monitoring results and being difficult to meet the strict requirements of the complex industrial environment for motor operation status monitoring and control. Summary of the Invention

[0006] The purpose of the present invention is to provide a method for analyzing operation data based on motor control to solve the above technical problems.

[0007] To solve the above technical problems, the following solution is adopted in this scheme: A method for analyzing operation data based on motor control includes the following steps: Step S1: Obtain and segment motor sensing flow information: Based on multiple sensors arranged on the motor, real-time collect motor operation data to form continuous motor sensing flow information; segment the continuous motor sensing flow information to obtain multiple sensing flow information segments.

[0008] Segmented processing is to convert continuous streaming data into smaller chunks of data that are easier to analyze and process, facilitating better capture of the changes in the motor's operating state over different time periods and providing a basis for subsequent description of the operating state and fault diagnosis.

[0009] Step S2: Mine the operating state description vector of the motor and identify the pending sequence: For each segment of sensor streaming information, conduct characterization information mining to obtain the operating state description vector of the motor sensor streaming information; based on the operating state description vector, identify the pending motor operating state sequence corresponding to the motor sensor streaming information.

[0010] Characterization information mining is to extract key features from each segment of sensor streaming information that can reflect the motor's operating state, while the identification of the operating state sequence is to preliminarily determine the motor's operating state over different time periods.

[0011] Step S3: Determine the state-significant data points and the positioning interval: In the pending motor operating state sequence, find the segment of sensor streaming information with abnormal motor operating state, and find the state-significant data points in this segment of sensor streaming information; based on the state-significant data points, determine the positioning interval of the corresponding state-indicating information flow in the motor sensor streaming information. By determining the positioning interval, the key position of the motor operating state mutation can be more accurately located.

[0012] Step S4: Determine and extract the target operating state description vector: In the operating state description vector, extract the operating state sub-description vectors of all data points within the positioning interval to obtain the target operating state description vector of the state-indicating information flow.

[0013] Step S5: Optimize the pending sequence to obtain the final sequence: Based on the target operating state description vector, optimize the pending motor operating state sequence to obtain the final motor operating state sequence corresponding to the current motor sensor streaming information.

[0014] Through optimization, the accuracy and reliability of the motor operating state sequence are improved, making it closer to the actual operating state of the motor. The final motor operating state sequence can more accurately reflect the evolution of the motor's operating state and provide an important basis for the control and maintenance of the motor.

[0015] Step S6: According to the final motor operating state sequence, take control measures in a timely manner for abnormal states during subsequent motor operation to achieve precise control and ensure the motor resumes normal operation.

[0016] For further optimization, in step S1, motor operating data is collected, including current data, voltage data, temperature data, and speed data. The motor sensor streaming information contains multiple data points, and each data point contains a timestamp and corresponding four-dimensional feature data.

[0017] For further optimization, in step S1, the sliding window technique is used to cut the continuous motor sensing flow information into overlapping or non - overlapping sensing flow information segments S j , j ∈[1, N], where N represents dividing the data stream into N sensing flow information segments.

[0018] For further optimization, step S2 specifically includes: Step S2.1, single - segment feature extraction: For each sensing flow information segment S j , through the state monitoring algorithm, the time - series features are extracted to generate the initial operating state description vector V j = v j1 , v j2 , v j3 , v j4 , v ji represents the operating state feature value of the j th type of sensor monitoring data within the i th sensing flow information segment; i ∈[1, 4], representing current, voltage, temperature, and rotational speed in sequence.

[0019] Step S2.2, global state splicing: Splice the initial operating state description vectors of all sensing flow information segments in chronological order to obtain the global operating state description vector V = [V1, V2, …, V N of the current motor sensing flow information.

[0020] Step S2.3, initial state sequence recognition: Determine the motor operating state corresponding to the initial operating state description vector of each sensing flow information segment, so as to obtain the undetermined motor operating state sequence Q = q 1, q 2, …, q N corresponding to the global operating state description vector V, where q j is the motor operating state corresponding to the initial operating state description vector of the j th sensing flow information segment, such as "normal operation", "overload operation", and "fault operation". This sequence is a preliminary inference of the motor operating state and provides a basis for subsequent precise analysis.

[0021] For further optimization, step S3 specifically includes: Step S3.1. Detect mutation conditions and identify significantly status data points: For the sensor flow information segments with abnormal motor operating states in the to-be-determined motor operating state sequence, calculate the operating state descriptor vectors for each data point. Based on the operating state descriptor vectors, estimate the probabilities of the instantaneous motor characteristics corresponding to these data points. Compare the probability of each data point with a preset threshold, and mark the data points that exceed the preset threshold d k as significantly status data points, and record the mutation occurrence positions corresponding to these data points idx ( k ).

[0022] Step S3.2. Form mutation condition groups and locate intervals: Group all the significantly status data points into G mutation condition groups; determine the starting mutations s g and ending mutations e g corresponding to each mutation condition group, and respectively determine the data points corresponding to the starting mutations s g and ending mutations e g in the motor sensor flow information. Move the position of the starting mutation data point idx ( s g ) forward by ΔL1 points, and move the position of the ending mutation data point idx ( e g ) backward by ΔL2 points to form the target data point coverage range idx ( s g ) - ΔL1, idx ( e g ) + ΔL2] of this mutation condition group, which is denoted as the location interval of the status indication information flow corresponding to this mutation condition group; Use the same method to obtain the location intervals of the status indication information flows corresponding to all mutation condition groups.

[0023] For further optimization, in step S4, for each location interval, classify the data points included in idx ( s g ) - ΔL1, idx ( e g ) + ΔL2] into the candidate sensor monitoring data point set, and the operating state descriptor vectors of all data points in the candidate sensor monitoring data point set constitute the target operating state description vector V g .

[0024] For further optimization, step S5 includes: Step S5.1: Determine the motor operating state corresponding to each operating state description sub-vector in the target operating state description vector, and form a target motor operating state subsequence q for the corresponding positioning interval. g ′.

[0025] Step S5.2: Use the target motor operating state subsequence q g ′ to replace the motor operating state corresponding to the positioning area in the sensing flow information segment with abnormal operating state in the to-be-determined motor operating state sequence Q, and obtain the final motor operating state sequence Q = q 1, q 2, …, q g ′, …, q N corresponding to the current motor sensing flow information.

[0026] For further optimization, in the above step S2.1, for the sensing flow information segment S j ={x1, x2,......, x n}, perform characterization information mining to obtain the initial operating state description vector corresponding to each sensing flow information segment, where the data point x t = I ( t ), U( t ), T( t ), w( t )], including t the current, voltage, temperature and rotational speed values of the motor at a certain moment; specifically including: Step S2.1.1: Data preprocessing and dataset construction: Pre-collect a large amount of motor sensing flow information and segment it into several sensing flow information segments as samples.

[0027] Perform standardization processing on different types of sensing data in the sensing flow information segment respectively, so that each data feature has a similar scale range. For example, perform normalization operations on current, voltage, temperature and rotational speed data respectively to make their values fall within intervals such as [0, 1] or (-1, 1), which helps to improve the training efficiency and stability of the RNN network.

[0028] Divide the preprocessed large amount of sensing flow information segment data into a training set, a validation set and a test set according to 7:1.5:1.5.

[0029] Step S2.1.2: Build a motor operating state monitoring model based on RNN and CTC networks: Determine the input layer: According to the feature dimensions of the current, voltage, temperature and rotational speed data of each sensing flow information segment n, determine that the number of neurons in the input layer is 4×n. Assume that there are n sampling points for each data feature in the sensing flow information segment, then the number of neurons in the input layer can be set to 4×n, which respectively correspond to the sampling points of 4 types of sensor data, and are used to receive the multi-dimensional data input of each information segment.

[0030] Determine the RNN hidden layer: Select the RNN cell type as the long short-term memory network LSTM or the gated recurrent unit GRU, and determine the number of layers in the hidden layer and the number of neurons in each layer.

[0031] LSTM or GRU cells can effectively process the long-term dependencies in sequential data and play an important role in capturing the time series features in the sensing flow information segment. The hidden layer is generally 3 - 5 layers, which can better extract deep features. The number of neurons in each layer can be adjusted according to experience or through experiments, and common choices are 64, 128, 256, etc.

[0032] Determine the CTC network layer: Add a CTC (Connectionist Temporal Classification) network layer after the RNN hidden layer, which is used to calculate the probability distribution between the input sequence and the output label sequence, enabling the model to directly classify or describe the state of the variable-length sensing flow information segment without forcibly aligning the lengths of the input and output sequences, because the length of the initial running state description vector may be different from the length of the input sensing data segment.

[0033] Determine the output layer: Determine that the number of neurons in the output layer is 4 according to the dimension of the initial running state description vector. Each neuron corresponds to the predicted value of a state feature and is used to output the initial running state description vector.

[0034] Step S2.1.3, Model training: Use the training set in step S2.1.1 to train the model. Adopt the CTC loss function as the training objective function of the model, and optimize the model parameters by minimizing the CTC loss, so that the model can accurately predict the corresponding initial running state description vector according to the input sensing flow information segment; adopt the stochastic gradient descent SGD or Adam optimization algorithm to update the network parameters.

[0035] The Adam optimization algorithm, due to its characteristic of adaptive learning rate adjustment, can better balance the convergence speed and stability, and usually shows good results when training RNN-based models, and is preferred.

[0036] During the training process, the training data is input into the constructed model batch by batch. The output results are calculated through forward propagation, and then the loss value is calculated using the CTC loss function. Next, the network parameters are updated through backpropagation and optimization algorithms. After each training epoch, the validation set is used to validate the model, and the performance metrics of the model on the validation set, such as loss value, accuracy, etc., are evaluated. According to the validation results, the hyperparameters of the model (such as learning rate, number of neurons in the hidden layer, etc.) and the parameter update strategy during the training process are adjusted to prevent overfitting and underfitting until the performance of the model on the validation set reaches the preset training stop condition.

[0037] Step S2.1.4, Real-time output of the initial operating state description vector of the motor: The current, voltage, temperature, and rotational speed data of the motor collected by the sensor are segmented and processed according to the same preprocessing process as the training data. The real-time preprocessed sensor streaming information segment data is input into the trained state monitoring model based on the RNN and CTC networks, and the model outputs the corresponding initial operating state description vector.

[0038] Based on the initial operating state description vector, it is possible to understand in real time whether the current operating state of the motor is normal and whether there are any abnormal or fault signs, so as to take corresponding control measures in a timely manner to ensure the stable operation of the motor. At the same time, as new data is continuously input, the model can update the monitoring results of the operating state in real time, realizing continuous tracking and monitoring of the motor operating state.

[0039] For further optimization, in step S2.3, calculate the motor operating state corresponding to the initial operating state description vector of each sensor streaming information segment, so as to obtain the undetermined motor operating state sequence Q corresponding to the global operating state description vector V, which specifically includes: Step S2.3.1, Set the mapping rule: According to the physical meaning of the motor operating state and expert knowledge, set the rule for mapping the initial operating state description vector to the specific motor operating state.

[0040] Step S2.3.2, State determination: For the initial operating state description vector V of each sensor streaming information segment j , make a determination according to the above mapping rule to obtain the corresponding motor operating state q j ; Step S2.3.3, Arrangement of the motor operating state sequence: Arrange the motor operating states corresponding to the initial operating state description vectors in the operating state description vector V in chronological order to obtain the undetermined motor operating state sequence Q = q 1, q 2,…, q N .

[0041] For further optimization, in step S3.1, for the sensor streaming information segment with abnormal motor operating states in the to-be-determined motor operating state sequence, calculate the operating state descriptor vectors for each data point in the corresponding sensor streaming information segment, and based on the operating state descriptor vectors, estimate the possibility of the instantaneous motor characteristics corresponding to these data points. Specifically, it includes: Step S3.1.1, extraction of the operating state descriptor vector, including: Step S3.1.1.1, adjustment and training of the model: Based on the above state monitoring model of the RNN-CTC network, for the need to obtain the operating state descriptor vector, adjust the output part of the model. That is, add a sub-vector generation layer between the RNN hidden layer and the CTC network layer, and the number of neurons in this layer is determined to be 4 according to the expected dimension of the operating state descriptor vector.

[0042] Train the adjusted model. The training data includes the sensor streaming information segment data of the motor in normal and abnormal operating states collected in advance. After training, the adjusted model can accurately output the initial operating state description vector and also generate an operating state descriptor vector with certain representational significance for each data point. For example, the multi-task learning method can be adopted to combine the initial operating state description vector prediction task with the sub-vector generation task for each data point, and through joint training, the model can learn the global operating state characteristics and the characteristics of local data points simultaneously.

[0043] Step S3.1.1.2, extraction of the sub-vector: After the adjusted model is trained, for the sensor streaming information segment with abnormal motor operating states in the to-be-determined motor operating state sequence, preprocess the information segment data according to the model input requirements and then input it into the adjusted model; during the forward propagation process of the model, after extracting the sequence features through the RNN hidden layer, the operating state descriptor vectors for each data point in the sensor streaming information segment can be obtained through the newly added sub-vector generation layer.

[0044] Step S3.1.2, estimation of the instantaneous motor characteristics, including: Step S3.1.2.1, definition and annotation of the characteristic categories: Clearly define the 4 types of situations included in the instantaneous motor characteristics, including sudden current change, voltage fluctuation, sudden temperature rise, and abnormal sudden change in speed, and perform annotation definitions for these 4 situations.

[0045] According to historical data or expert experience, determine the specific characteristics and judgment criteria for each characteristic category. For example, a sudden current change may be characterized by the current value exceeding a certain threshold within a short time and having a relatively fast change rate, etc.

[0046] Step S3.1.2.2, construction and training of the classification model: Collect pre - labeled motor operation data samples containing information on different transient characteristic categories. This sample covers various normal and abnormal transient characteristic situations of the motor, and use this sample as the dataset for training the classification model; each sample element includes the operation status description sub - vector corresponding to the data point and the corresponding transient characteristic category label.

[0047] Select classification algorithms such as random forest, support vector machine, logistic regression or deep neural network to construct a classification model; use the operation status description sub - vectors in the sample as input features and the corresponding transient characteristic category labels as output targets to train the classification model, and minimize the loss function (such as cross - entropy loss, etc.) by adjusting the classification model parameters, so that the classification model can accurately predict the corresponding motor transient characteristic category according to the operation status description sub - vector.

[0048] Step S3.1.2.3, Possibility estimation: For the operation status description sub - vector of each data point in the sensing flow - type information segment to be predicted, input it into the trained classification model; The classification model will output the probabilities of this sub - vector corresponding to each transient characteristic category, that is, the possibilities.

[0049] Compared with the prior art, the present invention has the following beneficial effects: 1) By comprehensively analyzing multi - dimensional data such as current, voltage, temperature and speed, the present invention comprehensively captures the motor operation status information. Compared with the method that only relies on a single type of sensing data, it can detect the motor operation status more accurately.

[0050] 2) When dealing with the sudden state of the motor, by determining the significant data points of the state and the positioning interval of the state - indicating information flow, it can accurately locate the beginning and end of a complete sudden state, accurately grasp the evolution process of the motor operation status, and provide strong support for timely and accurate control decisions.

[0051] 3) Controlling the motor based on the final motor operation status sequence can achieve precise management of the motor operation status. Adjusting the control strategy in a timely manner according to the actual operation status of the motor can improve the safety, stability and working efficiency of the motor, reduce the occurrence of faults, extend the service life of the motor, and has significant practical application value. Brief Description of the Drawings

[0052] Figure 1 It is a schematic diagram of the implementation process of the operation data analysis method based on motor control described in the invention; Figure 2 It is a hardware schematic diagram of a computer system described in the present invention. Detailed Embodiments

[0053] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be further elaborated in detail below with reference to the accompanying drawings and embodiments. The described embodiments should not be construed as limitations on this application. All other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of this application.

[0054] Embodiment 1: As Figure 1 shown, a method for analyzing operation data based on motor control includes the following steps: Step S100: Obtain and segment motor sensing flow information: Based on multiple sensors arranged on the motor, collect motor operation data in real time to form continuous motor sensing flow information; segment the continuous motor sensing flow information to obtain multiple sensing flow information segments.

[0055] In this embodiment, collect the sensing flow information of a motor on an industrial production line during 100 consecutive seconds of operation, as shown in Table 1. For the sake of brevity, only partial data is provided. This motor is mainly used to drive the conveyor belt. During these 100 seconds, the conveyor belt will experience working conditions such as starting, stable operation, short-term overload, resuming stable operation, and stopping. Through sensors installed on the motor, collect the current data, voltage data, temperature data, and rotational speed data of the motor at a frequency of once per second.

[0056] Table 1 Sensing flow information of the motor during 100 consecutive seconds of operation

[0057] Segment the current motor sensing flow information every 10 seconds as a time period to obtain 10 sensing flow information segments, as shown in Table 2.

[0058] Table 2 Sensing flow information segments of motor operation

[0059] Step S200: Mine the motor operation state description vector and identify the pending sequence: Conduct characterization information mining on each sensing flow information segment to obtain the operation state description vector of the motor sensing flow information; based on the operation state description vector, identify the pending motor operation state sequence corresponding to the motor sensing flow information.

[0060] In this embodiment, use the state monitoring algorithm based on RNN, which includes a CTC network. The input layer receives the current, voltage, temperature, and rotational speed data of each sensing flow information segment, the hidden layer extracts and transforms these data, and the output layer outputs the initial operation state description vector.

[0061] Taking the first sensing flow information segment (1 - 10 seconds) as an example, its data is shown in Table 3, including 10 data points, and each data point includes four values: current, voltage, temperature, and rotational speed.

[0062] Data of the first sensing flow information segment in Table 3

[0063] Input these data into the state monitoring algorithm. After calculation, the initial operating state description vector corresponding to this sensing flow information segment is V1 = [0.2, 0.8, 0.1, 0.9], where each element represents the operating state characteristic value of the motor in terms of current, voltage, temperature, and rotational speed.

[0064] In the same way, process the other 9 sensing flow information segments to obtain their respective corresponding initial operating state description vectors, as shown in Table 4.

[0065] Table 4 Initial operating state description vectors corresponding to 10 sensing flow information segments

[0066] Combine the above 10 initial operating state description vectors in sequence to obtain the operating state description vector V corresponding to the current motor sensing flow information: [0.2, 0.8, 0.1, 0.9], [0.22, 0.81, 0.11, 0.92], [0.23, 0.82, 0.12, 0.93], ... [0.19, 0.8, 0.1, 0.88] According to the physical meaning of the motor operating state and expert knowledge, set the rules for mapping the initial operating state description vector to the specific motor operating state.

[0067] In this embodiment, three operating states of the motor are predefined: normal operation, overload operation, and fault operation, corresponding to three standard vectors. Among them, the standard vector corresponding to normal operation is [0.2, 0.8, 0.1, 0.9]; the standard vector corresponding to overload operation is [0.5, 0.7, 0.3, 0.6]; the standard vector corresponding to fault operation is [0.8, 0.2, 0.6, 0.2].

[0068] In this embodiment, the set rule is: by calculating the initial operating state description vector V corresponding to each segment in the operating state description vector j ​​The similarity (using Euclidean distance) with these three standard vectors is used to determine the sequence of the to-be-determined motor operating states.

[0069] Taking the initial operating state description vector [0.2, 0.8, 0.1, 0.9] corresponding to the first sensing flow information segment as an example, calculate its Euclidean distances from the three standard vectors: The Euclidean distance from the normal operating standard vector: ; The Euclidean distance from the overload operating standard vector: ; The Euclidean distance from the fault operating standard vector: .

[0070] Since the Euclidean distance from the normal operating standard vector is the smallest, the to-be-determined motor operating state corresponding to this segment is normal operation.

[0071] Calculating in the same way for the other 9 sensing flow information segments, the sequence Q of the to-be-determined motor operating states obtained is: [normal operation, normal operation, normal operation, overload operation,..., normal operation].

[0072] In other embodiments, the rule can be set as: when the values of each dimension of the motor initial operating state description vector are within the normal range, the current dimension value is between [I norm_low , I norm_high , the voltage dimension value is between [U norm_low , U norm_high , the temperature dimension value is between [T norm_low , T norm_high , and the rotational speed dimension value is between [R norm_low , R norm_high .

[0073] Correspondingly, the overload state is: the current dimension value exceeds I norm_high , and the rotational speed dimension value shows a downward trend. The locked-rotor state is: the rotational speed dimension value is close to zero and the current dimension value surges. The undervoltage state is: the voltage dimension value is lower than U low . Compare the initial operating state description vector V j corresponding to each sensing flow information segment with the above ranges to obtain the corresponding motor operating state.

[0074] Step S300: Determine the state significant data points and the positioning interval: Search for the sensing flow information segments with abnormal motor operating states in the sequence of the to-be-determined motor operating states, and find the state significant data points in the sensing flow information segments; based on the state significant data points, determine the positioning interval of the corresponding state indication information flow in the motor sensing flow information.

[0075] In this embodiment, according to the above-mentioned to-be-determined motor operating state sequence, it is determined that the motor operating state of the 5th sensing flow information segment is abnormal. The operating state descriptor vectors of each data point in the corresponding sensing flow information segment are obtained by using the adjusted model. The logistic regression classification model is used to estimate the likelihood values of the instantaneous characteristics of the motor corresponding to each sensing monitoring data point. Specifically, by inputting 10 operating state descriptor vectors in the 5th sensing flow information segment, the probabilities of the motor corresponding to each sensing monitoring data point being in the normal operating, overload operating, and fault operating states are output. In other embodiments, classification models can be constructed by selecting algorithms such as random forest, support vector machine, or deep neural network classification algorithms.

[0076] In this embodiment, taking the sensing monitoring data point at the 51st second as an example, the corresponding operating state descriptor vector [0.4, 0.75, 0.2, 0.7] is input into the logistic regression model, and the probabilities of the motor corresponding to this data point being in the normal operating, overload operating, and fault operating states are obtained as follows: the normal operating probability P1 = 0.3; the overload operating probability P2 = 0.6; the fault operating probability P3 = 0.1.

[0077] Set the threshold of the overload operating probability to 0.5. When the overload operating probability corresponding to a certain sensing monitoring data point exceeds 0.5, it is considered that a mutation condition has occurred at the position where this data point is located.

[0078] By judging the likelihood values of all sensing monitoring data points, it is found that the overload operating probabilities corresponding to the sensing monitoring data points from the 51st to the 53rd second all exceed 0.5. Therefore, these three time points are determined as the positions where mutations occur. The corresponding sensing monitoring data points are recorded as state significant data points. Since there is only one group of consecutive mutation occurrence positions from the 51st to the 53rd second, a mutation condition group is obtained.

[0079] According to the mutation occurrence positions, it is determined that the start of the mutation condition is the 51st second and the end of the mutation condition is the 53rd second in the mutation condition group.

[0080] Determine the serial number of the state significant data point corresponding to the start of the mutation condition in the current motor sensing flow information idx ( s g ) is 51.

[0081] Determine the serial number of the state significant data point corresponding to the end of the mutation condition in the current motor sensing flow information idx ( e g ) is 53.

[0082] Set the first preset value ΔL1 to 1 and the second preset value ΔL2 to 1. Subtract the starting mutation condition data point sequence number from the first preset value to obtain the target starting data point sequence number of 50. Add the ending mutation condition data point sequence number to the second preset value to obtain the target ending data point sequence number of 54. Thus, the positioning interval of the state indication information flow corresponding to this mutation condition group is from the 50th to the 54th second, that is, the target data point coverage range is from the 50th to the 54th second. Therefore, the candidate set of sensing monitoring data points is the sensing monitoring data points at the 50th, 51st, 52nd, 53rd, and 54th seconds, and the corresponding sensing monitoring data points are shown in Table 5.

[0083] Table 5 Sensing Monitoring Data from the 50th to the 54th Seconds

[0084] Step S400: Determine the extraction of the target operating state description vector: Extract the operating state sub-description vectors of all data points within the positioning interval from the operating state description vector to obtain the target operating state description vector of the state indication information flow.

[0085] In this embodiment, the operating state sub-description vectors corresponding to the 50th to the 54th seconds in the operating state description vector are respectively: [0.3, 0.78, 0.15, 0.85] (the 50th second); [0.4, 0.75, 0.2, 0.7] (the 51st second); [0.45, 0.72, 0.25, 0.65] (the 52nd second); [0.42, 0.73, 0.22, 0.68] (the 53rd second); [0.35, 0.76, 0.18, 0.8] (the 54th second). These sub-description vectors form the set of operating state sub-description vectors.

[0086] Take the set of operating state sub-description vectors as the target operating state description vector corresponding to the state indication information flow. Then the target operating state description vector corresponding to the state indication information flow is: [0.3, 0.78, 0.15, 0.85], [0.4, 0.75, 0.2, 0.7], [0.45, 0.72, 0.25, 0.65], [0.42, 0.73, 0.22, 0.68], [0.35, 0.76, 0.18, 0.8] .

[0087] Step S500: Optimize the undetermined sequence to obtain the final sequence: Based on the target operating state description vector, optimize the undetermined motor operating state sequence to obtain the final motor operating state sequence corresponding to the current motor sensing flow information. ​

[0088] In this embodiment, the Euclidean distances between each sub-description vector in the target operating state description vector and three standard vectors (normal operation, overload operation, and fault operation) are calculated to determine the operating state of the information flow motor corresponding to each operating state sub-description vector.

[0089] Taking the sub-description vector [0.4, 0.75, 0.2, 0.7] corresponding to the 51st second as an example: The Euclidean distance from the normal operation standard vector: ; The Euclidean distance from the overload operation standard vector: ; The Euclidean distance from the fault operation standard vector: .

[0090] Since the Euclidean distance from the overload operation standard vector is the smallest, the operating state of the motor corresponding to the 51st second in the continuous motor sensing flow information is overload operation.

[0091] In the same way, the operating states of the motor corresponding to the 50th - 54th seconds are determined as follows: {50th second: normal operation; 51st second: overload operation; 52nd second: overload operation; 53rd second: overload operation; 54th second: normal operation}.

[0092] According to the positioning interval, the operating states of the motor corresponding to the 50th - 54th seconds are combined to form the target motor operating state subsequence q g ′: {normal operation (50th second), overload operation (51st - 53rd seconds), normal operation (54th second)}.

[0093] Using the target motor operating state subsequence q g ′, replace the motor operating state corresponding to the positioning area in the sensing flow information segment with abnormal operating state in the to-be-determined motor operating state sequence Q to obtain the final motor operating state sequence Q′ = q 1, q 2,…,q g ′,…, q N .

[0094] In this embodiment, replace the motor operating state corresponding to the 50th - 54th seconds in the 5th sensing flow information segment of the to-be-determined motor operating state sequence Q with the target motor operating state subsequence q g′, the final motor operation status sequence Q′ is: [Normal operation, normal operation, normal operation,..., normal operation (at the 50th second), overload operation (from the 51st to the 53rd second), normal operation (at the 54th second), normal operation,..., normal operation].

[0095] Step S600: According to the final motor operation status sequence Q′, it is found that the motor has an overload operation status from the 51st to the 53rd second. To avoid damage to the motor due to long-term overload, the following control measures are taken: When it is monitored that the motor enters the overload operation status (at the 51st second), immediately reduce the load of the conveyor belt, such as reducing the number of items on the conveyor belt. At the same time, increase the heat dissipation measures of the motor, such as turning on an additional cooling fan, to reduce the temperature of the motor. Continuously monitor the operation status of the motor. After the motor resumes normal operation (at the 54th second), gradually restore the normal load of the conveyor belt.

[0096] Embodiment 2: A computer system includes a memory and a processor. The memory stores a computer program that can run on the processor. When the processor executes the program, it implements some or all of the steps in the above method.

[0097] Figure 2 This is a schematic diagram of the hardware entity of a computer system provided in this embodiment. The hardware entity of the computer system 1000 includes: a processor 1001 and a memory 1002. Among them, the memory 1002 stores a computer program that can run on the processor 1001. When the processor 1001 executes the program, it implements the steps in the method of any of the above embodiments.

[0098] The memory 1002 stores a computer program that can run on the processor. The memory 1002 is configured to store instructions and applications executable by the processor 1001, and can also cache data to be processed or already processed by the processor 1001 and each module in the computer system 1000 (for example, image data, audio data, voice communication data, and video communication data), and can be implemented by flash memory (FLASH) or random access memory (Random Access Memory, RAM).

[0099] When the processor 1001 executes the program, it implements the steps of the operation data analysis method based on motor control in any of the above items. The processor 1001 generally controls the overall operation of the computer system 1000.

[0100] This application embodiment provides a computer storage medium. The computer storage medium stores one or more programs. The one or more programs can be executed by one or more processors to implement the steps of the operation data analysis method based on motor control in Embodiment 1.

[0101] As described above, it is only the implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application can easily think of changes or substitutions, which should all be covered within the protection scope of the present application.

Claims

1. A method for analyzing operation data based on motor control, characterized in that, Including the following steps: Step S1: Obtain and segment motor sensing flow information: Based on multiple sensors arranged on the motor, collect motor operation data in real time to form continuous motor sensing flow information; Segment the continuous motor sensing flow information to obtain multiple sensing flow information segments; Step S2: Mine the motor operation state description vector and identify the pending sequence: Perform characterization information mining on each sensing flow information segment to obtain the operation state description vector of the motor sensing flow information; Based on the operation state description vector, identify the pending motor operation state sequence corresponding to the motor sensing flow information; Step S3: Determine the state significant data points and the positioning interval: Search for the sensing flow information segments with abnormal motor operation states in the pending motor operation state sequence, and find the state significant data points in this sensing flow information segment; Based on the state significant data points, determine the positioning interval of the corresponding state indication information flow in the motor sensing flow information; Step S4: Determine and extract the target operation state description vector: Extract the operation state sub-description vectors of all data points within the positioning interval from the operation state description vector to obtain the target operation state description vector of the state indication information flow; Step S5: Optimize the pending sequence to obtain the final sequence: Based on the target operation state description vector, optimize the pending motor operation state sequence to obtain the final motor operation state sequence corresponding to the current motor sensing flow information; Step S6: According to the final motor operation state sequence, take control measures in a timely manner for abnormal states during subsequent motor operation to ensure that the motor resumes normal operation.

2. The method for analyzing operation data based on motor control according to claim 1, wherein In the said step S1, collecting motor operation data includes current data, voltage data, temperature data and speed data. The motor sensing flow information contains multiple data points, and each data point contains a timestamp and corresponding four-dimensional feature data.

3. The method for analyzing operation data based on motor control according to claim 2, wherein In the step S1, the sliding window technique is adopted to cut the continuous motor sensing flow information stream into overlapping or non-overlapping sensing flow information segments S j , j ∈[1, N], where N represents that the continuous motor sensing flow information is divided into N sensing flow information segments in total.

4. The operation data analysis method based on motor control according to claim 3, wherein The said step S2 specifically includes: Step S2.1, single-segment feature extraction: For each sensing flow information segment S j , perform time-series feature extraction through a state monitoring algorithm to generate an initial operating state description vector V j = v j1 , v j2 , v j3 , v j4 , v ji indicating the operating state eigenvalue of the j th type of sensing monitoring data within the i th sensing flow information segment; i ∈ [1, 4], representing current, voltage, temperature, and rotational speed in sequence; Step S2.2, Global State Concatenation: Concatenate the initial running state description vectors of all sensor streaming information segments in chronological order to obtain the global running state description vector V = [V1, V2, …, V N ; Step S2.

3. Initial state sequence recognition: Determine the motor operating state corresponding to the initial operating state description vector of each sensing flow information segment, so as to obtain the to-be-determined motor operating state sequence Q = q 1, q 2, …, q N corresponding to the global operating state description vector V, where q j is the motor operating state corresponding to the initial operating state description vector of the j th sensing flow information segment.

5. The operating data analysis method based on motor control according to claim 4, wherein The said step S3 specifically includes: Step S3.1, detecting mutation conditions and identifying significantly status data points: For the sensor flow information segments with abnormal motor operating states in the to-be-determined motor operating state sequence, calculate the operating state descriptor vectors for each data point. Based on the operating state descriptor vectors, estimate the probabilities of the corresponding instantaneous motor characteristics for these data points, compare the probability of each data point with a preset threshold, and d k mark the data points exceeding the preset threshold as significantly status data points, and record the mutation occurrence positions corresponding to these data points idx ( k ) Step S3.2, forming mutation working condition groups and positioning intervals: Classify all significantly abnormal data points into G mutation working condition groups; determine the starting mutation s g and ending mutation e g for each mutation working condition group, and respectively determine the data points corresponding to the starting mutation s g and ending mutation e g in the motor sensing flow information. Push the position of the starting mutation data point idx ( s g ) forward by ΔL1 points, and extend the position of the ending mutation data point idx ( e g ) backward by ΔL2 points to form the target data point coverage range idx ( s g ) - ΔL1, idx ( e g ) + ΔL2] of this mutation working condition group, which is denoted as the positioning interval of the state indication information flow corresponding to this mutation working condition group; Using the same method, obtain the positioning intervals of the state indication information flows corresponding to all mutation working condition groups.

6. The method for analyzing operation data based on motor control according to claim 5, wherein In step S4, for each positioning interval, idx ( s g ) - ΔL1, idx ( e g ) + ΔL2], the data points included are classified into the candidate sensing and monitoring data point set, and the operation state descriptor vectors of all data points in the candidate sensing and monitoring data point set constitute the target operation state description vector V g .

7. The method for analyzing operation data based on motor control according to claim 6, wherein The said step S5 includes: Step S5.1: Determine the motor operating state corresponding to each operating state description sub-vector in the target operating state description vector, and form a target motor operating state subsequence q g ′; Step S5.

2. Use the target motor operating state subsequence q g ′ to replace the motor operating state corresponding to the positioning area in the sensing flow information segment with abnormal operating state in the to-be-determined motor operating state sequence Q, and obtain the final motor operating state sequence Q′ = q 1, q 2, …, q g ′, …, q N .

8. The operation data analysis method based on motor control according to claim 7, wherein In the step S2.1, for the sensing flow information segment S j ={x1,x2,......,x n}, perform characterization information mining to obtain the initial operating state description vector corresponding to each sensing flow information segment, where the data point x t = I ( t ),U( t ),T( t ),w( t )], including t the current, voltage, temperature and rotational speed values of the motor at the moment; Specifically includes: Step S2.1.1, Data preprocessing, constructing a data set: Pre-collect a large amount of motor sensing flow information and segment it to form several sensing flow information segments as samples; Standardize different types of sensing data in the sensing flow information segments respectively to make the data features have similar scale ranges; Then divide the preprocessed sensing flow information segment data into a training set, a validation set and a test set according to 7:1.5:1.5; Step S2.1.2, Build a motor operation state monitoring model based on RNN and CTC networks: Determine the input layer: Based on the characteristic dimensions of the current, voltage, temperature, and rotational speed data of each sensing flow information segment n , determine that the number of neurons in the input layer is 4×n; Determine the RNN hidden layer: Select the RNN unit type as the long short-term memory network LSTM or the gated recurrent unit GRU, and determine the number of layers of the hidden layer and the number of neurons in each layer; Determine the CTC network layer: Add a CTC network layer after the RNN hidden layer to calculate the probability distribution between the input sequence and the output label sequence, enabling the model to directly classify or describe the state of variable-length sensing streaming information segments without forcibly aligning the lengths of the input and output sequences; Determine the output layer: Determine that the number of neurons in the output layer is 4 according to the dimension of the initial operating state description vector; Step S2.1.3, Model training: Use the training set in step S2.1.1 to train the model. Adopt the CTC loss function as the training objective function of the model, and optimize the model parameters by minimizing the CTC loss, so that the model can accurately predict the corresponding initial operating state description vector according to the input sensing streaming information segment; Use the stochastic gradient descent SGD or Adam optimization algorithm to update the network parameters; During the training process, input the training data in batches into the constructed model, calculate the output result through forward propagation, then calculate the loss value using the CTC loss function, and then update the network parameters through backpropagation and the optimization algorithm; After each training cycle, use the validation set to verify the model and evaluate the performance metrics of the model on the validation set until the performance of the model on the validation set reaches the preset training stop condition; Step S2.1.4, Real-time output of the initial operating state description vector of the motor: Segment the current, voltage, temperature, and speed data of the motor collected by the sensor and process them according to the same preprocessing process as the training data. Input the real-time preprocessed sensing streaming information segment data into the trained state monitoring model based on RNN and CTC networks, and the model outputs the corresponding initial operating state description vector.

9. The operation data analysis method based on motor control according to claim 8, wherein In step S2.3, determine the motor operating state corresponding to the initial operating state description vector of each sensing streaming information segment, so as to obtain the undetermined motor operating state sequence Q corresponding to the global operating state description vector V, specifically including: Step S2.3.1, Set the mapping rule: According to the physical meaning of the motor operating state and expert knowledge, set the rule for mapping the initial operating state description vector to the specific motor operating state; Step S2.3.2, Status Determination: For the initial running state description vector V of each sensing flow information segment j , make a determination according to the above mapping rules to obtain the corresponding motor running state q j ; Step S2.3.3, motor operating state sequence arrangement: Arrange the motor operating states corresponding to each initial operating state description vector in the operating state description vector V in chronological order to obtain a pending motor operating state sequence Q = q 1, q 2,…, q N .

10. The method for analyzing operation data based on motor control according to claim 9, wherein In step S3.1, for the sensing streaming information segments with abnormal motor operating states in the undetermined motor operating state sequence, find the operating state description sub-vectors of each data point in the corresponding sensing streaming information segment, and based on the operating state description sub-vectors, estimate the possibility of the instantaneous characteristics of the motor corresponding to these data points, specifically including: Step S3.1.1, Extraction of the operating state description sub-vector, including: Step S3.1.1.1, Adjust the model and train: Based on the state monitoring model based on RNN and including the CTC network constructed in claim 8, add a sub-vector generation layer between the RNN hidden layer and the CTC network layer; The number of neurons in this layer is determined to be 4 according to the expected dimension of the operating state description sub-vector; Train the adjusted model, where the training data includes pre-collected sensor streaming information segment data of the motor in normal and abnormal operating states; after training, the adjusted model can not only accurately output the initial operating state description vector, but also output the operating state description sub-vector of each data point; Step S3.1.1.2, extraction of the sub-vector of the sensor streaming information segment with abnormal motor operating state: After the adjusted model is trained, for the sensor streaming information segment data with abnormal motor operating state in the to-be-determined motor operating state sequence, after preprocessing, it is input into the adjusted model; during the forward propagation process of the adjusted model, after extracting the sequence features through the RNN hidden layer, the operating state description sub-vector of each data point in this sensor streaming information segment can be output through the newly added sub-vector generation layer; Step S3.1.2, estimation of the motor instantaneous characteristics, including: Step S3.1.2.1, definition and annotation of characteristic categories: clarify the 4 types of situations included in the instantaneous characteristics of the motor, including sudden current change, voltage fluctuation, sudden temperature rise, and abnormal sudden change in speed, and make annotation definitions for these 4 situations; and determine the specific characteristics and judgment criteria of each characteristic category according to historical data or expert experience; Step S3.1.2.2, construction and training of the classification model: Collect pre-annotated motor operating data samples containing information on different instantaneous characteristic categories, which cover various normal and abnormal instantaneous characteristic situations of the motor, and use this sample as the data set for training the classification model; each sample element includes the operating state description sub-vector of the corresponding data point and the corresponding instantaneous characteristic category label; Select a random forest, support vector machine, logistic regression, or deep neural network classification algorithm to construct a classification model; use the operating state description sub-vector in the sample as the input feature and the corresponding instantaneous characteristic category label as the output target to train the classification model so that the classification model can accurately predict the corresponding motor instantaneous characteristic category according to the operating state description sub-vector; Step S3.1.2.3, possibility estimation: Input the operating state description sub-vector of each data point in the sensor streaming information segment with abnormal motor operating state in the to-be-determined motor operating state sequence into the trained classification model, and the classification model outputs the probability of this operating state description sub-vector corresponding to each instantaneous characteristic category, that is, the possibility.

Images