Intelligent health monitoring method, system, device and medium for linear motion control mechanism

Abstract

The present invention belongs to the technical field of linear motion control mechanisms and discloses a method, system, device, and medium for intelligent health monitoring of linear motion control mechanisms. The method comprises fitting operating data to a first data feature based on a motor power formula; fitting the first data feature to a second image feature based on a motor angle signal; padding and resizing the second image feature to obtain a third image feature; and utilizing the third image feature in combination with transfer learning to fine-tune an existing convolutional neural network pre-trained model to achieve health monitoring. By converting the timing signal output by the motor during reciprocating motion of the linear motion control mechanism into a power-angle image representation, the present invention can fully utilize the limited motor output signal and effectively represent the different operating states of the linear motion control mechanism. The deep transfer learning architecture is employed to further improve the generalization performance and applicability of the established model on different data sets, as well as the accuracy of the model in practical applications.

Description

Technical Field

[0001] The present invention belongs to the technical field of linear motion control mechanisms, and in particular relates to a method, system, device and medium for intelligent health monitoring of a linear motion control mechanism. Background Art

[0002] Linear motion control mechanisms are one of the most important motion control systems and have been widely used in various industrial applications, including CNC machine tools, automated assembly lines, semiconductors, material handling, and automatic door systems. The accuracy of the motion control system directly determines the accuracy capability of the entire system. A healthy linear motion control mechanism can accurately respond to position controller commands. However, when system performance degrades, the loss of positioning accuracy will increase significantly. Therefore, it is very important and meaningful to achieve the self-perception capability of the linear motion control mechanism's health status. CN201911125048.4 discloses a linear motion system health monitoring method based on spatial domain information, which belongs to the field of electromechanical system health status monitoring. The spatial position signal of the linear motion system and the electrical signal of the motor are measured, and the measurement information based on the time domain is converted to the spatial domain. The local characteristic indicators of the local travel interval and the global characteristic indicators of the entire motion travel are calculated to determine whether they exceed the corresponding set thresholds. However, the judgment of the converted characteristic signal indicators is relatively simple and the generalization performance is relatively poor.

[0003] Through the above analysis, the problems and defects of the existing technology are as follows: the existing technology has a low utilization rate of the collected characteristic signals, it is difficult to effectively characterize the abnormal modes of the linear motion system, the rules for judging abnormal states are too simple, and the established model has poor generalization performance for other linear motion systems. Summary of the Invention

[0004] In response to the problems existing in the prior art, the present invention provides a method, system, device and medium for intelligent health monitoring of a linear motion control mechanism.

[0005] The present invention implements an intelligent health monitoring method for a linear motion control mechanism. Based on the analysis of the operation process of the linear motion control mechanism, the timing signal of the drive motor during the operation of the system, including current, speed, and angle, is converted into an image representation method. On this basis, system health monitoring and abnormality identification are performed. The specific principle is as follows: Figure 1 As shown in the figure: the operating data is fitted into the motor power signal according to the motor power formula; the power signal in a reciprocating motion is converted into a power-angle closed curve by combining the motor angle signal, and the power-angle closed curve is filled and regularized to obtain a power-angle image representation. Finally, health monitoring is achieved based on the convolutional neural network model and transfer learning algorithm.

[0006] For the intelligent health monitoring method of the linear motion control mechanism, the operating data is fitted into the first data feature according to the motor power formula; the first data feature is fitted into the second image feature according to the motor angle signal; the second image feature is filled and resized to obtain the third image feature; the third image feature is used in combination with transfer learning to fine-tune the existing convolutional neural network pre-training model to achieve health monitoring.

[0007] Furthermore, the second image feature obtained by fitting the first data feature to the angle signal includes: taking the angle signal as the x-axis and the first data feature as the y-axis, a second image feature is obtained from a reciprocating motion power-angle curve.

[0008] Furthermore, filling and resizing the second image feature includes: internally filling the obtained closed power-angle curve to obtain a black and white image, unifying the x-axis and y-axis lengths of all images, and finally unifying the pixel size of the final image.

[0009] Furthermore, a data representation method of converting the time series data into an image is used to calculate the motor power and fit the collected operating data into a first data feature;

[0010]

[0011] Where P is power in watts (W); T is torque in Newton-meters (N·m); N is motor speed in radians per second (rad / s); and 9549 is a constant. Since T = torque constant * current, where the torque constant is a fixed value that only affects the power amplitude and does not change the power curve, torque T can be equivalently replaced by motor current.

[0012] Furthermore, the use of the third image feature to train the classification model to obtain the health monitoring results of each sample in the data set includes: assigning index values ​​to normal and abnormal types as labels for each type, dividing the source domain data into training set: test set = 8:2 to form a modeling data set; using a convolutional neural network model, such as the AlexNet pre-trained model, to train the classification model to ensure that the model accuracy meets the requirements; adopting a model-based transfer learning strategy, using a small amount of data in the target domain to fine-tune the classification model to obtain the health monitoring results of the target domain samples.

[0013] Furthermore, the use of the third image feature to train the classification model and obtain the health monitoring results of each sample in the data set also includes: using cropping, rotation, and scaling data enhancement techniques to expand the samples in the source domain and the target domain; using FocalLoss to modulate the category-imbalanced data set; using part of the target domain data to perform a convolutional layer parameter freezing test on the initial health monitoring model, selecting the freezing method with the least impact on the target domain health monitoring accuracy to freeze the initial model parameters, and finally using a small amount of target domain data to fine-tune and optimize the initial model to obtain the final health monitoring model.

[0014] Furthermore, the first data feature obtained according to the motor power formula also includes: preprocessing the motor operation data, eliminating data with abnormal length, filling null values ​​and normalizing.

[0015] Another object of the present invention is to provide a computer device, which includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the intelligent health monitoring method of the linear motion control mechanism.

[0016] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the intelligent health monitoring method for a linear motion control mechanism.

[0017] Another object of the present invention is to provide a linear motion control mechanism intelligent health monitoring system for implementing the linear motion control mechanism intelligent health monitoring method, the linear motion control mechanism intelligent health monitoring system comprising:

[0018] The data acquisition unit removes data with abnormal length, fills in null values ​​and normalizes the collected data;

[0019] a data preprocessing unit, which fits the linear motion control mechanism operation data into a first data feature according to the motor power formula, fits the first data feature into a second image feature according to the motor angle signal, and fills and regularizes the second image feature to obtain a third image feature;

[0020] The health monitoring unit is used to use the third image feature to perform classification model training to obtain a health monitoring result for each sample in the test set.

[0021] In combination with the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solutions to be protected by the present invention are as follows:

[0022] First, in response to the technical problems existing in the prior art and the difficulty of solving them, this paper closely combines the technical solutions claimed by the present invention with the results and data from the research and development process, and presents a detailed and in-depth analysis of the technical problems solved by the present invention's technical solutions, as well as the innovative technical effects achieved by solving these problems. Specifically, the paper addresses the following challenges: How to achieve the required accuracy while controlling testing costs, i.e., without adding external sensors, and using only the data from the linear motion control mechanism itself, is a challenging problem. By converting the timing signals output by the motor during the reciprocating motion of the linear motion control mechanism into power-angle images, the present invention fully utilizes the available motor output signals to effectively characterize the different operating states of the linear motion control mechanism, thereby improving the accuracy of health monitoring. Furthermore, the present invention employs a deep transfer learning architecture, based on a convolutional neural network model, to identify different abnormal patterns of the linear motion control mechanism, improving the model's accuracy on a single dataset. Furthermore, a model-based transfer learning approach is used to enhance the generalization and applicability of the established model across different datasets. This technical solution can significantly improve the accuracy of health monitoring and abnormality identification of linear motion control mechanisms in practical application scenarios without the need for external sensors.

[0023] Second, considering the technical solution as a whole or from the perspective of the product, the technical effects and advantages of the technical solution to be protected by the present invention are specifically described as follows: the present invention only uses the angle, current, and speed signals output by the motor of the linear motion control mechanism itself, without the need for additional sensors, which can greatly reduce testing costs; by converting the timing signal output by the motor during the reciprocating motion of the linear motion control mechanism into a power-angle image representation, it can make full use of the only motor output signal and effectively represent the different operating states of the linear motion control mechanism; the deep transfer learning architecture is adopted to further improve the generalization performance and applicability of the established model on different data sets, thereby improving the accuracy of the model in practical applications.

[0024] Third, the technical solution of the present invention is also reflected in the following important aspects:

[0025] (1) The expected benefits and commercial value of the technical solution of the present invention after transformation are:

[0026] The present invention can be applied to linear motion control mechanisms where it is inconvenient to install external sensors, to achieve health monitoring and abnormality identification. It can detect abnormalities in the early stages of failure in the linear motion system, guide timely maintenance work, and avoid the subsequent serious consequences in production quality, personnel safety, etc. caused by the continued deterioration of the fault, such as loss of motion accuracy and abnormal shutdown.

[0027] (2) The technical solution of the present invention fills the technical gap in the industry at home and abroad: The present invention proposes a health monitoring method for linear motion control mechanisms based on image feature information and deep transfer learning, which improves the accuracy of abnormality recognition of linear motion systems and the generalization performance of the model under different data sets. It is the first time that this method has been proposed and applied in this field. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 This is a flow chart of the intelligent health monitoring method for a linear motion control mechanism provided by an embodiment of the present invention;

[0029] Figure 2 This is a schematic diagram of the intelligent health monitoring method for a linear motion control mechanism proposed in the present invention;

[0030] Figure 3 This is a flow chart of the intelligent health monitoring method for a linear motion control mechanism provided by an embodiment of the present invention.

[0031] Figure 4 The figure shows a process of converting a motor timing signal into a power-angle diagram according to an embodiment of the present invention. DETAILED DESCRIPTION

[0032] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0033] like Figure 1 As shown, the intelligent health monitoring method for a linear motion control mechanism provided by an embodiment of the present invention includes the following steps:

[0034] S101: Acquiring operating data of the linear motion control mechanism in normal and different fault states, specifically, current, rotation angle, and speed signal data of the motor when the linear motion control mechanism is in reciprocating motion in each state;

[0035] S102: Cleaning the collected motor operation data, removing data with abnormal lengths, filling null values, and normalizing the data; plotting a success rate-angle curve based on the total power signals of the linear motion control mechanism during forward and reverse motion according to the change in angle, and filling the interior of the obtained closed power-angle curve with color to obtain a black and white image;

[0036] S103: Migrating the initial health monitoring model of the source domain dataset to the target domain. After expanding the target domain data through operations such as cropping, rotation, and scaling, a convolutional layer parameter freezing test is performed on the initial health monitoring model using part of the target domain data. The freezing method that has the least impact on the health monitoring accuracy of the target domain is selected to freeze the initial model parameters. The initial model is then fine-tuned and optimized using a small amount of labeled data from the target domain to obtain the final health monitoring model.

[0037] S104: Use the remaining data in the target domain to test the fine-tuned health monitoring model to achieve high-precision diagnosis.

[0038] like Figure 2 、 Figure 3 As shown, the embodiment of the present invention provides an intelligent health monitoring method for a linear motion control mechanism based on image feature information and transfer learning, including the following steps:

[0039] Step 1, data collection.

[0040] The operation data of the linear motion control mechanism under normal and different fault states are obtained, specifically the current, rotation angle and speed signal data of the motor when the linear motion control mechanism moves back and forth in each state.

[0041] Step 2: Data preprocessing.

[0042] (1) Clean the collected motor operation data, remove data with abnormal length, fill in null values ​​and normalize the data;

[0043] (2) According to the knowledge of mechanism, the current and speed signals are converted into power signals. The power calculation formula of the motor is:

[0044]

[0045] Where P is power in watts (W); T is torque (or moment) in Newton-meters (N-m); N is motor speed in radians per second (rad / s); and 9549 is a constant.

[0046] (3) The total power signal of the linear motion control mechanism during forward and reverse motion is plotted as a success rate-angle curve according to the angle change, and the closed power-angle curve is filled with color to obtain a black and white image. At the same time, the x and y axis lengths of all images are unified according to the change of the screw angle signal, and finally the pixel size of the final image is unified.

[0047] Step 3: Model training.

[0048] (1) Different linear motion control mechanism datasets are divided into source domains and target domains. The obtained source domain images are expanded through data augmentation techniques such as cropping, rotation, and scaling, and then input into the AlexNet pre-trained model, where the AlexNet pre-trained model uses network parameters pre-trained with the Imagenet dataset;

[0049] (2) The weight parameters of the convolutional pooling layer that extracts the feature structure in the pre-trained model are transferred to this dataset to build a new model. The class imbalanced dataset is modulated using FocalLoss to reduce the differences caused by class imbalance. Finally, the prediction results are output to complete the construction of the initial health monitoring model of the source domain dataset.

[0050] (3) The initial health monitoring model of the source domain dataset is migrated to the target domain. After the target domain data is expanded through operations such as cropping, rotation, and scaling, part of the target domain data is used to perform a convolutional layer parameter freezing test on the initial health monitoring model: no freezing, freezing to convolutional layer 1, convolutional layer 2, convolutional layer 3, convolutional layer 4, and convolutional layer 5. The freezing method with the least impact on the health monitoring accuracy of the target domain is selected to freeze the initial model parameters. Finally, a small amount of labeled data from the target domain is used to fine-tune and optimize the initial model to obtain the final health monitoring model.

[0051] Step 4: Model testing.

[0052] The fine-tuned health monitoring model is tested using the remaining data from the target domain to achieve high-accuracy diagnosis.

[0053] The intelligent health monitoring system for a linear motion control mechanism provided by an embodiment of the present invention includes:

[0054] The data acquisition unit removes data with abnormal length, fills in null values ​​and normalizes the collected data;

[0055] a data preprocessing unit, which fits the linear motion control mechanism operation data into a first data feature according to the motor power formula, fits the first data feature into a second image feature according to the motor angle signal, and fills and regularizes the second image feature to obtain a third image feature;

[0056] The health monitoring unit is used to use the third image feature to perform classification model training to obtain a health monitoring result for each sample in the test set.

[0057] In order to prove the creativity and technical value of the technical solution of the present invention, this section provides application examples of the claimed technical solution on specific products or related technologies.

[0058] The experiment was carried out on an electric door test bench. The electric door is driven by a motor to drive the lead screw to open and close the door, which is a typical linear motion control mechanism.

[0059] Step 1: Get the running data of the motor when the screw is running.

[0060] Experimental data was collected from multiple test benches. Each test bench contained data from normal motor operation and five different abnormal conditions during the experiment. Feature data such as current, rotation angle, and speed were extracted from each type of motor operating data. The experimental data was then divided into source domain data (Dataset A) and target domain data (Datasets B and C).

[0061] Step 2: Data preprocessing.

[0062] (1) Clean the collected motor operation data, remove data with abnormal length, fill in null values ​​and normalize the data, and finally unify the sequence data length to 320;

[0063] (2) Based on the knowledge of the mechanism, convert the current and speed signals into power signals;

[0064] (3) All power signals are plotted as a success rate-angle curve according to the angle change, and the closed power angle curve is filled with color to obtain a black and white image. At the same time, the x and y axis lengths of all images are unified, and finally the pixel size of the final image is unified.

[0065] Attachment Figure 4 Schematic diagram of the process of converting motor timing signals into power-angle diagram.

[0066] Step 3: Model training.

[0067] (1) Different linear motion control mechanism datasets are divided into source domain and target domain. The images of source domain dataset A are expanded through data augmentation techniques such as cropping, rotation, and scaling, and then input into the AlexNet pre-trained model. The AlexNet pre-trained model uses the network parameters pre-trained with the Imagenet dataset.

[0068] (2) The weight parameters of the convolutional pooling layer that extracts the feature structure in the pre-trained model are transferred to this dataset to build a new model. The class imbalanced dataset is modulated using FocalLoss to reduce the differences caused by class imbalance. Finally, the prediction results are output to complete the construction of the initial health monitoring model for the source domain dataset A.

[0069] (3) The initial health monitoring model of the source domain dataset A is migrated to the target domain datasets B and C. After the target domain data is expanded through operations such as cropping, rotation, and scaling, part of the target domain data is used to perform a convolutional layer parameter freezing test on the initial health monitoring model. The freezing method that has the least impact on the target domain health monitoring accuracy is selected to freeze the initial model parameters. Finally, a small amount of labeled data from the target domain is used to fine-tune and optimize the initial model to obtain the final health monitoring model.

[0070] Table 1 Comparison of different frozen pre-training model parameters

[0071]

[0072]

[0073] Step 4: Model testing.

[0074] The health monitoring model obtained by testing the target domain dataset B and the remaining data of dataset C achieves high-precision diagnosis.

[0075] It should be noted that the embodiments of the present invention can be implemented by hardware, software, or a combination of software and hardware. The hardware portion can be implemented using dedicated logic; the software portion can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated design hardware. Those skilled in the art will appreciate that the above-mentioned devices and methods can be implemented using computer-executable instructions and / or contained in processor control code, for example, such as a carrier medium such as a disk, CD or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The devices and modules of the present invention can be implemented by hardware circuits such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., can also be implemented by software executed by various types of processors, or can be implemented by a combination of the above-mentioned hardware circuits and software, such as firmware.

[0076] Tables 2 and 3 show the comparative effects of the technical solutions of the present invention and other technical solutions in the embodiments, wherein Table 2 compares the effects of the deep learning model proposed by the present invention with traditional machine learning methods. As can be seen from Table 2, the traditional SVM-based classification method has the problem of unstable recognition accuracy, and the accuracy rate is significantly different from the MCC index. However, the results obtained after training using the image-based AlexNet pre-trained model achieved a higher accuracy rate, and the model training accuracy rate was close to the MCC index data.

[0077] Table 2 Comparison of results between traditional machine learning methods and deep learning methods

[0078]

[0079] Table 3 compares the effects of the deep migration model proposed in the present invention with two technical solutions that do not use transfer learning, where Strategy 1 uses new target domain data to directly test the trained deep learning model, and Strategy 2 uses target domain data to directly train the AlexNet pre-trained model, and then uses the bench data to test the accuracy of the model. Strategy 3 is the AlexNet pre-trained model proposed in the present invention that uses a small amount of target domain bench data to fine-tune the AlexNet pre-trained model trained based on the source domain dataset A based on transfer learning. As can be seen from Table 3, when the method model proposed in Strategy 1 is tested, the model accuracy is poor. In comparison, the fault recognition accuracy of the method model proposed in Strategy 2 gradually improves as the number of samples increases, and the accuracy is optimal when the initial total number of samples reaches 100, proving that it is necessary to use target domain data to fine-tune the model parameters. The results of Strategy 3 under different data settings are significantly better than those of Strategy 2, proving the superiority of the technical solution of the present invention.

[0080] Table 3 Comparison of effects of different initial sample numbers in different data sets

[0081]

[0082] The above description is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with this technical field within the technical scope disclosed by the present invention and within the spirit and principles of the present invention should be covered by the scope of protection of the present invention.

Claims

1. A method for intelligent health monitoring of a linear motion control mechanism, characterized in that: The method for intelligent health monitoring of a linear motion control mechanism fits operating data into a first data feature based on a motor power formula; fits the first data feature into a second image feature based on a motor angle signal; pads and resizes the second image feature to obtain a third image feature; and utilizes the third image feature in combination with transfer learning to fine-tune an existing convolutional neural network pre-trained model to implement health monitoring. The second image feature fitted by the first data feature and the rotation angle signal includes: taking the rotation angle signal as the x-axis and the first data feature as the y-axis, obtaining a second image feature from a reciprocating motion power-rotation angle curve; Filling and resizing the second image feature includes: filling the internal portion of the obtained closed power-angle curve to obtain a black and white image, unifying the x-axis and y-axis lengths of all images, and finally unifying the pixel size of the final image; A data representation method for converting time series data into an image, calculating motor power, and fitting the collected operating data into a first data feature; Where P is power in watts (W); T is torque in Newton meters (N·m); N is motor speed in radians per second (rad / s); and 9549 is a constant. Since T = torque constant * current, where the torque constant is a fixed value that only affects the power amplitude and does not change the trend of the power curve, torque T can be equivalently replaced by motor current. The method of using the third image feature to train a classification model to obtain a health monitoring result for each sample in the data set includes: assigning index values ​​to normal and abnormal types as labels for each type, dividing the source domain data into a training set:test set ratio of 8:2 to form a modeling data set; using a pre-trained convolutional neural network model to train the classification model to ensure that the model accuracy meets the requirements; and fine-tuning the classification model using a small amount of data in the target domain to obtain the health monitoring results of the target domain samples. Before using the third image feature to train the classification model and obtain the health monitoring results of each sample in the data set, the method also includes: using cropping, rotation, and scaling data enhancement techniques to expand the samples in the source domain and the target domain; using FocalLoss to modulate the category-imbalanced data set; using part of the target domain data to perform a convolutional layer parameter freezing test on the initial health monitoring model, selecting the freezing method that has the least impact on the target domain health monitoring accuracy to freeze the initial model parameters, and finally using a small amount of target domain data to fine-tune and optimize the initial model to obtain the final health monitoring model.

2. The intelligent health monitoring method for a linear motion control mechanism according to claim 1, wherein: Before obtaining the first data feature according to the motor power formula, the method further includes: preprocessing the motor operation data, eliminating data with abnormal length, filling null values ​​and normalizing.

3. A computer device, characterized in that: The computer device includes a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the intelligent health monitoring method for a linear motion control mechanism according to any one of claims 1 to 2.

4. A computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the intelligent health monitoring method for a linear motion control mechanism according to any one of claims 1 to 2.

5. A linear motion control mechanism intelligent health monitoring system implementing the linear motion control mechanism intelligent health monitoring method according to any one of claims 1 to 2, characterized in that: The linear motion control mechanism intelligent health monitoring system includes: The data acquisition unit removes data with abnormal length, fills in null values ​​and normalizes the collected data; a data preprocessing unit, which fits the linear motion control mechanism operation data into a first data feature according to the motor power formula, fits the first data feature into a second image feature according to the motor angle signal, and fills and regularizes the second image feature to obtain a third image feature; The health monitoring unit is used to use the third image feature to perform classification model training to obtain a health monitoring result for each sample in the test set.

Images