Magnetorheological polishing abnormal state monitoring system and method based on neural network

Abstract

The present invention belongs to the field of real-time monitoring technology for optical processing, and in particular, relates to a neural network-based system and method for monitoring abnormal conditions in magnetorheological polishing. The method comprises: S1: acquiring a set of ribbon data from a complete magnetorheological polishing process using a displacement sensor; S2: extending each set of ribbon data so that the operating duration of each set is an integer multiple of the current sampling period; S3: re-dividing each set of ribbon data based on the sampling period T to obtain m training groups, and labeling each training group to obtain a training set; S4: inputting the training set into a neural network structure for training to obtain a neural network model; S5: inputting the real-time measured ribbon data to be monitored into the neural network model for processing until the neural network model's monitoring results indicate normality. The present invention establishes a relationship between ribbon changes during the polishing process and the processing status, thereby achieving automated and accurate monitoring of abnormal conditions in magnetorheological polishing.

Description

Technical Field

[0001] The present invention belongs to the technical field of real-time monitoring of optical processing, and in particular relates to a system and method for monitoring abnormal states of magnetorheological polishing based on a neural network. Background Art

[0002] Magnetorheological polishing (MRP) is a representative optical processing technique widely used in the field due to its advantages, such as minimal subsurface damage and high removal efficiency. MR fluid forms a Bingham flow under the influence of a magnetic field, generating shear forces that remove material. Therefore, the stability of the MR polishing ribbon indirectly reflects the stability of the MR polishing process.

[0003] During long, multi-round processes, magnetorheological polishing systems can experience anomalies, such as iron filings blocking the pipes. These anomalies can lead to reduced polishing efficiency at best, or even non-convergent machining results at worst. Therefore, the automation level of real-time monitoring of anomalies during multi-round, long-term magnetorheological polishing needs to be improved.

[0004] The Chinese invention patent entitled "Neural Network-Based Magnetorheological Polishing Removal Function Prediction Device and Method" (publication number CN117556345A, publication date March 12, 2024) predicts a real-time removal function using a ribbon profile. Its purpose is to detect real-time polishing removal efficiency, but its monitoring and judgment capabilities for different abnormal conditions are weak, and it is unable to automatically handle abnormal conditions. Summary of the Invention

[0005] In view of this, the present invention aims to provide a magnetorheological polishing abnormal state monitoring system and method based on a neural network to solve the problem that the existing technology has weak monitoring and judgment capabilities for different abnormal states and cannot automatically process abnormal states. The present invention establishes a relationship between the ribbon changes in the polishing process and the processing state by establishing a neural network prediction structure, thereby realizing automatic and accurate magnetorheological polishing abnormal state monitoring.

[0006] To achieve the above object, the technical solution created by the present invention is implemented as follows:

[0007] A neural network-based magnetorheological polishing abnormal state monitoring system, comprising:

[0008] Displacement sensor for obtaining A set of ribbon data under the complete magnetorheological polishing process;

[0009] The data processing module is used to set a sampling period, regroup the n groups of ribbon data based on the sampling period, and add labels to the grouped data to obtain a training set;

[0010] The neural network module is used to input the training set into the neural network structure for training to obtain the neural network model;

[0011] The processing control module is used to input the real-time measured data of the ribbon to be monitored into the neural network model for processing. If the monitoring result of the neural network model shows normal, the workpiece is subjected to magnetorheological polishing based on the current processing grinding head. Otherwise, the current processing grinding head is processed and the real-time measured data of the ribbon to be monitored is input into the neural network model for processing until the monitoring result of the neural network model shows normal.

[0012] Furthermore, in the data processing module, the sampling period is set to be greater than the inverse of the sampling frequency of the displacement sensor itself, and the sampling period is set to be an integer multiple of the inverse of the sampling frequency of the displacement sensor itself.

[0013] A method for monitoring abnormal conditions of magnetorheological polishing based on a neural network is applicable to a system for monitoring abnormal conditions of magnetorheological polishing based on a neural network. The method specifically comprises the following steps:

[0014] S1: Displacement sensor collection A complete set of ribbon data under the magnetorheological polishing process , is the ribbon profile at the tth second under the nth set of complete magnetorheological polishing process;

[0015] S2: Extend each set of ribbon data so that the working duration of each set of ribbon data is an integer multiple of the current sampling period;

[0016] S3: Set a sampling period T, re-divide each group of ribbon data based on the sampling period T to obtain m training groups, and add labels to each training group to obtain a training set;

[0017] S4: Input the training set into the neural network structure for training to obtain a neural network model;

[0018] S5: The real-time measured data of the ribbon to be monitored is input into the neural network model for processing. If the monitoring result of the neural network model shows normal, the workpiece is subjected to magnetorheological polishing based on the current processing grinding head. Otherwise, the current processing grinding head is processed, and the real-time measured data of the ribbon to be monitored is input into the neural network model for reprocessing until the monitoring result of the neural network model shows normal.

[0019] Furthermore, in step S2, the specific operation of extending each set of ribbon data is: copying the data of the last sampling period of each set of ribbon data.

[0020] Furthermore, in step S3, the specific operation of adding labels to each training group is as follows:

[0021] If the magnetic streamer polishing operation reflected by the current training group is in normal state, then label 0 is added to the current training group; otherwise, label 1 is added to the current training group.

[0022] Repeat the above steps until all training groups are labeled.

[0023] Furthermore, in step S4, the neural network structure includes a first LSTM layer, a second LSTM layer and a fully connected layer connected in sequence.

[0024] Furthermore, the loss function N used to train the neural network structure is:

[0025] ;

[0026] in, is the true label, is the predicted probability value, and m is the number of training samples in the training set.

[0027] Compared with the prior art, the present invention can achieve the following beneficial effects:

[0028] This invention provides a neural network-based system and method for monitoring abnormal conditions in magnetorheological polishing. To improve the automation level of the magnetorheological polishing system and enhance its unmanned operation capabilities, the invention processes magnetorheological ribbon profile change data collected by displacement sensors during previous processes and uses this data set to train the neural network. When a new magnetorheological polishing monitoring task arises, simply input the magnetorheological polishing ribbon profile change data obtained by the displacement sensors into the network to quickly determine whether the process status is abnormal. Compared to traditional manual monitoring methods, the proposed method offers significant automation advantages. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The accompanying drawings, which constitute part of the present invention, are intended to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are intended to explain the present invention and do not constitute an undue limitation of the present invention. In the accompanying drawings:

[0030] Figure 1 A schematic flow chart of a method for monitoring abnormal conditions of magnetorheological polishing based on a neural network according to an embodiment of the present invention;

[0031] Figure 2 A schematic diagram of the training process of the neural network structure described in the embodiment of the present invention.

[0032] Description of reference numerals:

[0033] 1. First LSTM layer; 2. Second LSTM layer; 3. Fully connected layer. DETAILED DESCRIPTION

[0034] 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 with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and do not constitute a limitation of the present invention.

[0035] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments can be combined with each other.

[0036] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. In addition, the terms "first", "second" and the like are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, features defined as "first", "second" and the like may explicitly or implicitly include one or more of the features. In the description of the present invention, unless otherwise specified, "multiple" means two or more.

[0037] In the description of the present invention, it should be noted that, unless otherwise expressly specified or limited, the terms "installed," "connected," and "connected" should be understood in a broad sense. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections or electrical connections; they can refer to direct connections or indirect connections through an intermediate medium; and they can refer to internal connections between two components. Those skilled in the art can understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0038] The present invention will be described in detail below with reference to the accompanying drawings and in combination with embodiments.

[0039] like Figure 1 As shown, the present invention provides a magnetorheological polishing abnormal state monitoring system based on a neural network, comprising:

[0040] Displacement sensor for obtaining A set of ribbon data under the complete magnetorheological polishing process;

[0041] The data processing module is used to set a sampling period, regroup the n groups of ribbon data based on the sampling period, and add labels to the grouped data to obtain a training set;

[0042] The neural network module is used to input the training set into the neural network structure for training to obtain the neural network model;

[0043] The processing control module is used to input the real-time measured data of the ribbon to be monitored into the neural network model for processing. If the monitoring result of the neural network model shows normal, the workpiece is subjected to magnetorheological polishing based on the current processing grinding head. Otherwise, the current processing grinding head is processed and the real-time measured data of the ribbon to be monitored is input into the neural network model for processing until the monitoring result of the neural network model shows normal.

[0044] It should be noted that to prevent abnormal fluctuations during magnetorheological polishing that could affect machining accuracy and convergence efficiency, existing magnetorheological polishing processes often require manual oversight, leaving room for improvement in automation. To enhance the automation level and unattended operation of magnetorheological polishing systems, the present invention collects data on polishing ribbon changes during the magnetorheological polishing process to train a neural network architecture, thereby providing alerts and automated processing for abnormal conditions that occur during the process.

[0045] In some embodiments, in the data processing module, the sampling period is set to be greater than the inverse of the sampling frequency of the displacement sensor itself, and the sampling period is set to be an integer multiple of the inverse of the sampling frequency of the displacement sensor itself.

[0046] like Figure 2 As shown, the present invention proposes a method for monitoring abnormal conditions of magnetorheological polishing based on a neural network, which is applicable to a magnetorheological polishing abnormal condition monitoring system based on a neural network. The method specifically includes the following steps:

[0047] S1: Displacement sensor collection A complete set of ribbon data under the magnetorheological polishing process , is the ribbon profile at the tth second under the nth set of complete magnetorheological polishing process;

[0048] S2: Extend each set of ribbon data so that the working duration of each set of ribbon data is an integer multiple of the current sampling period;

[0049] The displacement sensor itself has a sampling period, and the current sampling period here is the sampling period of the displacement sensor itself.

[0050] S3: Set a sampling period T, re-divide each group of ribbon data based on the sampling period T to obtain m training groups, and add labels to each training group to obtain a training set;

[0051] S4: Input the training set into the neural network structure for training to obtain a neural network model;

[0052] S5: The real-time measured data of the ribbon to be monitored is input into the neural network model for processing. If the monitoring result of the neural network model shows normal, the workpiece is subjected to magnetorheological polishing based on the current processing grinding head. Otherwise, the current processing grinding head is processed, and the real-time measured data of the ribbon to be monitored is input into the neural network model for reprocessing until the monitoring result of the neural network model shows normal.

[0053] In some embodiments, in step S2, the specific operation of extending each set of ribbon data is: copying the data of the last sampling period of each set of ribbon data.

[0054] In some embodiments, in step S3, the specific operation of adding labels to each training group is:

[0055] If the magnetic streamer polishing operation reflected by the current training group is in normal state, then label 0 is added to the current training group; otherwise, label 1 is added to the current training group.

[0056] Repeat the above steps until all training groups are labeled.

[0057] In some embodiments, in step S4, the neural network structure includes a first LSTM layer 1, a second LSTM layer 2 and a fully connected layer 3 connected in sequence.

[0058] In some embodiments, the loss function N used to train the neural network structure is:

[0059] ;

[0060] in, is the true label, is the predicted probability value, and m is the number of training samples in the training set.

[0061] Example 1

[0062] The collected data is processed using a displacement sensor to obtain multiple sets of data on the changes in the magnetorheological polishing ribbon during the complete polishing process. The abnormal and normal parts of the change data (ribbon data) are labeled to create a data set. This data set is used to train the neural network structure. When the next processing process is carried out, the trained neural network model can judge the magnetorheological polishing status by inputting real-time ribbon profile data.

[0063] For example, the specific steps of monitoring the abnormal state of magnetorheological polishing in the present invention include:

[0064] S1, collected through displacement sensor A complete set of ribbon data for the processing flow ,in, Representative The ribbon profile at the tth second when the group completes the processing flow;

[0065] S2, set the sampling period to seconds, that is, each complete processing flow The second polishing ribbon data is integrated into a set of sampling data. At this time, it should be noted that the sampling period should not be less than the sampling period of the displacement sensor itself. The sampling period of the displacement sensor itself is 0.002 seconds, and it should be a positive integer multiple of the sampling frequency of the displacement sensor itself. Considering that the processing time of different groups of tasks may not be the same as the sampling period In order to facilitate the subsequent training of the neural network structure, the ribbon data is first extended, that is, the data of the last cycle of each set of ribbon data is copied so that the overall processing time is an integer multiple of the sampling period. For example, for the first The data before processing is shown in formula (1), and after processing is shown in formula (2):

[0066] (1);

[0067] (2);

[0068] (3);

[0069] in, Indicates the time when data collection starts. Indicates the time when data collection ends.

[0070] According to the sampling period The first Group ribbon data into The group, due to being in the last The data length of the group and the previous The data length of the group may not be consistent, copy the last second of the ribbon The processed data set is shown in formula (3). Finally, the data set is based on Each small group of data obtained from the ribbon data is labeled, 0 represents normal and 1 represents abnormal;

[0071] S3, the dataset (a total of The training set is divided into a training set and a test set with a ratio of 8:2. The neural network structure is then trained using the training set. The total number of training sets is groups, each group was treated as above The ribbon change matrix in time, specifically, the number of data points collected by the displacement sensor itself at one time is , then each sampling period The size of the ribbon change matrix inside is ,like Figure 2 As shown in the figure, the network used for training is a two-layer LSTM plus a fully connected layer 3. Since the LSTM network has a strong fitting ability for time series signals, the Adam optimizer is used for model training, and the loss function is binary cross entropy:

[0072] (4);

[0073] in, is the true label (0 or 1), is the probability value predicted by the model, is the number of training set samples;

[0074] S4. When real-time monitoring of abnormal state of magnetorheological polishing is required, By inputting the ribbon change data within seconds, the processing status at that time can be obtained. If there is an abnormality in the processing, the output of the neural network model will become 1. At this time, the abnormal status will be transmitted to the processing control unit, and the processing control unit will perform an emergency lifting and stopping operation on the processing grinding head to prevent the processing abnormality from affecting the final processing result.

[0075] Although the present invention requires a large initial dataset, once the neural network computational model is built and trained, monitoring of abnormal conditions during the magnetorheological polishing process will be highly accurate. Therefore, compared to traditional manual monitoring, the present invention has a high degree of automation.

[0076] It should be understood that the various forms of the processes shown above can be used to reorder, add, or delete steps. For example, the steps described in the present disclosure can be performed in parallel, sequentially, or in a different order, as long as the desired results of the technical solutions disclosed in the present disclosure can be achieved. This is not limited herein.

[0077] The above specific embodiments do not limit the scope of protection of the present invention. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions may be made based on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention are intended to be included within the scope of protection of the present invention.

Claims

1. A neural network-based magnetorheological polishing abnormal state monitoring system, characterized by: include: Displacement sensor for obtaining A set of ribbon data under the complete magnetorheological polishing process; The data processing module is used to set a sampling period, regroup the n groups of ribbon data based on the sampling period, and add labels to the data of each group obtained after grouping to obtain a training set; The neural network module is used to input the training set into the neural network structure for training to obtain the neural network model; The processing control module is used to input the real-time measured data of the ribbon to be monitored into the neural network model for processing. If the monitoring result of the neural network model shows normal, the workpiece is subjected to magnetorheological polishing based on the current processing grinding head. Otherwise, the current processing grinding head is processed and the real-time measured data of the ribbon to be monitored is input into the neural network model for processing until the monitoring result of the neural network model shows normal.

2. The neural network-based magnetorheological polishing abnormal state monitoring system according to claim 1 is characterized by: In the data processing module, the sampling period is set to be greater than the inverse of the sampling frequency of the displacement sensor itself, and the sampling period is set to be an integer multiple of the inverse of the sampling frequency of the displacement sensor itself.

3. A method for monitoring abnormal state of magnetorheological polishing based on neural network, characterized in that: The method is applicable to the neural network-based magnetorheological polishing abnormal state monitoring system according to claim 1 or claim 2, and specifically comprises the following steps: S1: Use displacement sensor to obtain A complete set of ribbon data under the magnetorheological polishing process , is the ribbon profile at the tth second under the nth set of complete magnetorheological polishing process; S2: Extend each set of ribbon data so that the working duration of each set of ribbon data is an integer multiple of the current sampling period; S3: Set a sampling period T, re-divide each group of ribbon data based on the sampling period T to obtain m training groups, and add labels to each training group to obtain a training set; S4: Input the training set into the neural network structure for training to obtain a neural network model; S5: The real-time measured data of the ribbon to be monitored is input into the neural network model for processing. If the monitoring result of the neural network model shows normal, the workpiece is subjected to magnetorheological polishing based on the current processing grinding head. Otherwise, the current processing grinding head is processed, and the real-time measured data of the ribbon to be monitored is input into the neural network model for reprocessing until the monitoring result of the neural network model shows normal.

4. The method for monitoring abnormal state of magnetorheological polishing based on neural network according to claim 3 is characterized in that: In step S2, the specific operation of extending each set of ribbon data is: copying the data of the last sampling period of each set of ribbon data.

5. The method for monitoring abnormal state of magnetorheological polishing based on neural network according to claim 3 is characterized in that: In step S3, the specific operations of adding labels to each training group are as follows: If the magnetic streamer polishing operation reflected by the current training group is in normal state, then label 0 is added to the current training group; otherwise, label 1 is added to the current training group. Repeat the above steps until all training groups are labeled.

6. The method for monitoring abnormal state of magnetorheological polishing based on neural network according to claim 3 is characterized in that: In step S4, the neural network structure includes a first LSTM layer, a second LSTM layer and a fully connected layer connected in sequence.

7. The method for monitoring abnormal state of magnetorheological polishing based on neural network according to claim 3 is characterized in that: The loss function N used to train the neural network structure is: ； in, is the true label, is the predicted probability value, and m is the number of training samples in the training set.

Images