Fault identification and detection method for box-type wire storage box turnover robot
By constructing a fault detection model based on multidimensional feature vectors and stacking ensemble learning algorithm, the problems of variability and complexity in fault identification and detection of box-type wire storage and turning robots are solved, achieving high-precision real-time fault identification and early warning, and ensuring the continuity of the production line and the stability of the equipment.
Patent Information
- Application Number
- CN202411278045.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-12
- Publication Date
- 2026-03-13
AI Technical Summary
Existing technologies suffer from inconsistencies, complexity, and insufficient rapid response capabilities in fault identification and detection of box-type wire storage and turning robots, which affect the continuity and stability of the production line.
A fault detection and identification model is constructed using multidimensional feature vectors and stacking ensemble learning algorithms. Robot operating status data is acquired through vibration sensors, preprocessed and labeled, and features are selected using the probability density distance method. Preliminary predictions are made by combining extreme gradient boosting, K-nearest neighbors, convolutional neural networks and deep neural networks, and the final results are optimized through logistic regression to achieve real-time fault identification.
It achieves high-precision identification and detection of faults in the box-flipping robot, with a prediction accuracy of over 95%, timely warning of potential faults, ensuring production continuity and equipment stability, and reducing unplanned downtime.
Smart Images
Figure CN121658856A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of tobacco equipment fault diagnosis technology, and more specifically, to a fault identification and detection method for a box-type tobacco storage and turning robot. Background Technology
[0002] The industrial robot's box-turning and feeding process is crucial in cigarette production, serving as the core equipment for supplying tobacco from the tobacco processing workshop to the packaging workshop. Since box-turning and feeding is a cyclical process, any malfunction occurring within a cycle can cause the entire production line to halt, impacting overall production efficiency. Therefore, a method is needed to achieve real-time monitoring and fault detection of the box-turning robot's operating status to prevent malfunctions, reduce downtime, and ensure the continuity and stability of production.
[0003] Because customized technologies and equipment are mostly used in periodic material feeding production processes, existing fault identification and detection methods will vary in application. Furthermore, the working environment of the crate-flipping robot is complex and variable, potentially affected by various factors such as temperature, humidity, and mechanical wear. Secondly, the robot's failure modes are extensive, covering a range from minor performance degradation to complete functional failure, thus complicating fault detection. In addition, due to the continuous requirements of the production line, the fault diagnosis system must have rapid response capabilities to take timely measures and minimize downtime.
[0004] In summary, there is a need to provide an improved technical solution that addresses the shortcomings of the existing technology. Summary of the Invention
[0005] The purpose of this application is to provide a fault identification and detection method for a box-type wire storage and flipping robot, which can detect faults in the box-type wire storage and flipping robot and provide early warnings to meet the needs of continuous production lines.
[0006] This application specifically provides a fault identification and detection method for a box-type wire storage and turning robot, including the following steps:
[0007] S1. Obtain vibration data of different operating states of the box-type wire storage and flipping robot, and establish a pre-made database after adding tags to the vibration data;
[0008] S2. Extract the multi-dimensional feature vector of the vibration data in the prefabricated database, and construct a fault detection and identification model based on the multi-dimensional feature vector;
[0009] S3. Based on the fault detection and identification model, monitor the operating status of the box-type wire storage and turning robot in real time and issue early warnings.
[0010] In one feasible embodiment, step S1 includes at least the following:
[0011] S11. Collect vibration data of the box-type wire storage and flipping robot; the vibration data includes at least vibration data under normal conditions and vibration data under known fault conditions;
[0012] S12. Preprocessing vibration data: Select vibration data from one working cycle as the processing object, and preprocess the processing object to obtain a pre-made database of vibration data.
[0013] S13. Add tags to the vibration data in the prefabricated database.
[0014] In one feasible approach, in step S11, a vibration sensor is installed at the bottom of the box-type wire storage and flipping robot. The vibration sensor is used to acquire vibration data of the base platform of the box-type wire storage and flipping robot under different operating conditions in real time.
[0015] In one feasible approach, in step S13, the vibration data in the prefabricated database is tagged according to the type of operating state: the data of the fault state is assigned a first tag, and the data of the normal operating state is assigned a second tag.
[0016] In one feasible approach, in step S2, the multidimensional feature vector includes at least a variety of time-domain statistical features and a variety of frequency-domain features.
[0017] In one feasible approach, in step S2, the multidimensional feature vectors are sorted based on their importance.
[0018] In one feasible embodiment, step S2 includes at least the following:
[0019] S21. Extract the multidimensional feature vectors of vibration data in the prefabricated database, and extract and filter the multidimensional feature vectors;
[0020] S22. Based on the extracted and filtered feature vectors, a fault identification and detection model is constructed using the stacking ensemble learning algorithm.
[0021] S23. Train and fit the fault identification and detection model;
[0022] S24. Determine and store the fault identification and detection model.
[0023] In one feasible approach, in step S21, evaluation indices S(d) of multiple feature vectors are obtained sequentially, and the feature vectors are sorted in descending order according to the numerical value of S(d) to complete the selection of feature vectors.
[0024] In one implementable manner, step S23 includes at least the following:
[0025] S231. Divide the vibration dataset in the prefabricated database into a training set and a test set;
[0026] S232. Construct a stacking ensemble learning model and initialize relevant model parameters, including the selection of base learners and meta-learners;
[0027] S233. Train a stacking ensemble learning model using the data from the training set;
[0028] S234. When the model reaches the iteration completion standard, use the test set to verify the accuracy, and output the predicted labels after completing the evaluation of the model performance.
[0029] In one feasible embodiment, step S3 includes at least the following:
[0030] S31. Acquire real-time data: Collect vibration data of the base platform of the box-flipping robot in real time, and divide the vibration data into multiple data blocks according to the predetermined working cycle;
[0031] S32. Identify real-time faults based on partitioned data blocks: Input the data blocks generated in each work cycle into the fault identification and detection model and determine the output discrimination result;
[0032] S33. Based on the judgment results, perform fault warning and response.
[0033] Compared with the prior art, the beneficial effects of this application are as follows:
[0034] In the technical solution of this application, an ensemble learning method is employed to achieve high-precision identification and detection of faulty operating states of a crate-flipping robot. The fault identification and detection model analyzes vibration signal data for each cycle and extracts key time-frequency domain features to construct an input vector. The feature vector is then filtered using a probability density distance method to identify the most representative features. Based on these features, a stacking ensemble learning model is constructed. This model performs initial predictions by stacking multiple basic learners and optimizes the prediction results using a meta-learner, thereby significantly improving the overall prediction performance of the model.
[0035] The fault identification and detection model provided in this application can monitor the operating status of the box-turning robot in real time and accurately warn of potential faults, thereby ensuring the continuity of production and the stability of the equipment. The fault identification and detection model provided in this application has a prediction accuracy rate of over 95%, and can promptly warn of equipment faults. Based on the predicted fault location, fault time, and shift information of the fault occurrence, maintenance personnel can conduct in-depth investigations and repairs in conjunction with business conditions, thereby effectively preventing production interruptions, reducing resource waste, and reducing the number of unplanned downtimes during production. Attached Figure Description
[0036] Figure 1 This is a flowchart of a fault identification and detection method for a box-type wire storage and flipping robot according to an embodiment of the present invention.
[0037] Figure 2 This is a flowchart illustrating the establishment of a fault identification and detection model in the fault identification and detection method of the box-type wire storage and flipping robot according to an embodiment of the present invention.
[0038] Figure 3 This is a flowchart of step S23 in the fault identification and detection method of the box-type wire storage and flipping robot of the present invention. Detailed Implementation
[0039] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings. These embodiments are for illustrative purposes only and are not intended to limit the scope of the invention.
[0040] In the description of this invention, it should be noted that the terms "center," "longitudinal," "lateral," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.
[0041] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.
[0042] Furthermore, in the description of this invention, unless otherwise stated, "a plurality of" means two or more.
[0043] See Figures 1 to 3 This application specifically provides a fault identification and detection method for a box-type wire storage and turning robot, including the following steps:
[0044] S1. Acquire vibration data of the box-type wire storage and flipping robot under different operating states, and add labels to the vibration data to obtain a pre-made database. Store the pre-made database to facilitate subsequent feature extraction and model training.
[0045] S2. Construct a fault detection and identification model.
[0046] S3. By monitoring the vibration data of the box-type wire storage and flipping robot in real time and inputting the vibration data into the fault detection and identification model, the operating status of the box-type wire storage and flipping robot can be monitored and an early warning can be issued.
[0047] In one feasible embodiment, step S1 includes at least the following:
[0048] S11. Collect vibration data.
[0049] A vibration sensor is installed at the bottom of the box-type wire storage and flipping robot. The vibration sensor is used to acquire vibration data of the base platform of the box-type wire storage and flipping robot in real time under different operating conditions.
[0050] It should be noted that the vibration data includes at least vibration data under normal conditions and vibration data under known fault conditions.
[0051] S12, Preprocess vibration data.
[0052] Vibration data from one working cycle is selected as the processing object, and the processing object is preprocessed to obtain a pre-built database of vibration data.
[0053] In one feasible approach, the preprocessing includes at least: imputing missing values and detecting and processing outliers.
[0054] S13. Add labels to the vibration data.
[0055] Add tags to the data in the prefabricated database according to the type of operating status: assign the first tag to the data in the fault state and assign the second tag to the data in the normal operating state.
[0056] Specifically, in this embodiment, based on the historical vibration data of the box-flipping robot, vibration data in a fault state is labeled with label 0, and vibration data in a normal state is labeled with label 1.
[0057] In one feasible approach, such as Figure 2 As shown, step S2 includes at least the following steps:
[0058] S21. Extract the multidimensional feature vectors of vibration data in the prefabricated database, normalize and standardize the multidimensional feature vectors, and complete the feature selection.
[0059] S22. Based on the extracted and filtered features, a fault identification and detection model is constructed using the stacking ensemble learning algorithm.
[0060] S23. Train and fit the fault identification and detection model based on data from the pre-built database. By training with data, optimize the model parameters and improve the fault identification accuracy of the fault identification and detection model.
[0061] S24. Deploy the fault identification and detection model: Integrate the trained fault identification and detection model into the programmable logic controller (PLC) to achieve real-time communication and fault detection with the box-turning robot on the production line.
[0062] In one feasible approach, the normalization process is based on a probability density distance method to rank the importance of multidimensional feature vectors in order to remove redundant features.
[0063] In one feasible approach, the multidimensional feature vector includes at least a variety of time-domain statistical features and a variety of frequency-domain features. The time-domain statistical features include at least the maximum value, maximum absolute value, minimum value, average value, peak-to-peak value, absolute average / rectified average value, root mean square value, root square amplitude, standard deviation, kurtosis / kurtosis, skewness, margin factor, waveform factor, impulse factor, peak factor, and kurtosis factor. The frequency-domain features include at least the frequency center, frequency center frequency, root mean square frequency, and root square frequency.
[0064] In one feasible approach, step S21, the screening features specifically includes the following:
[0065] Sixteen time-domain statistical features and four frequency-domain features of the vibration signal are extracted to form a multidimensional feature vector {x1, x2, ..., x...} D}, where D is the sample dimension. The feature vector is Z-score standardized according to formula (1).
[0066]
[0067] In the formula:
[0068] i is the sample number;
[0069] j represents the dimension.
[0070] The importance of feature vectors is ranked based on the probability density distance method. Specifically, an evaluation index S(d) is obtained for multiple feature vectors, and they are sorted in descending order according to the value of S(d) to select important features and filter out redundant feature vectors. The evaluation index S(d) is obtained according to formula (2):
[0071]
[0072] In the formula:
[0073] N is the number of samples;
[0074] D represents the sample dimension;
[0075] f(z ij ) is the probability density of the j-th dimension feature of sample i in the sample space;
[0076] f d ′(z ij ) is the estimated probability density of the d-th dimension feature of sample i in the sample space;
[0077] S(d) represents the evaluation index of the d-th dimension feature.
[0078] In one feasible manner, f d ′(z ij The process of determining ) is as follows: first, the d-th dimension feature of sample i is weighted and transformed, and then multiplied by the weight coefficient k, and then its probability density in the sample space is estimated.
[0079] In one feasible approach, such as Figure 3 As shown, step S23 includes at least the following:
[0080] S231. Divide the vibration dataset in the prefabricated database into a training set and a test set.
[0081] S232. Construct a stacking ensemble learning model and initialize relevant model parameters, including the selection of the base learner and meta-learner.
[0082] S233. Train a stacking ensemble learning model using the data from the training set.
[0083] S234. When the model reaches the iteration completion standard, use the test set to verify the accuracy, and output the predicted labels after evaluating the model performance.
[0084] Specifically, when the output prediction label is 0, it indicates that a fault has occurred in the current cycle; when the output prediction label is 1, it indicates that the current cycle is working normally.
[0085] In one feasible approach, in step S234, if the model has not met the iteration completion criteria, the model parameter settings are adjusted.
[0086] In one feasible approach, determining whether a model has reached the iteration completion criteria includes either the validation set accuracy no longer improving or the predetermined number of iterations being reached.
[0087] In one implementable manner, step S23 specifically includes the following steps:
[0088] S2321. Confirm the training set and test set.
[0089] In step S231, the vibration dataset in the prefabricated database is divided into a training set D. train ={(X train Y train )} and test set D test ={(X test Y test )}, where X represents the input feature vector and Y represents the output label.
[0090] S2322. Using the K-fold cross-validation method, the training set is divided into 4 subsets to obtain D. train1 D train2 D train3 D train4 Then, one set is used as the validation set and the other three sets are used as the training set.
[0091] S2323. Extreme gradient boosting (XGBoost), K-nearest neighbor (KNN), convolutional neural network (CNN) and deep neural network (DNN) are selected as the base learners for the first layer and named M1, M2, M3 and M4 respectively.
[0092] S2324. Train the base learners. For each base learner, train it using step S2322. Repeat step S2322 to obtain 3 sub-models and corresponding prediction values.
[0093] S2325. For any base learner M j And in step S2324, three predicted values are obtained, which are then merged to form dataset X. t ′ rainj .
[0094] S2326, Test set D test Input to each M j The average of the prediction results from the three sub-models is calculated to obtain X. t ′ estj .
[0095] S2327, Training the meta-learner, using X t ′ rainj and X t ′ estj The original dataset contains Y, which serves as the training and test set for the meta-learner logistic regression (LR) model, respectively. train and Y test As the corresponding output label.
[0096] S2328. Output prediction labels. A prediction label of 0 indicates a fault in the current cycle. A prediction label of 1 indicates that the current cycle is working normally.
[0097] By employing the Stacking ensemble learning framework, the model's prediction accuracy and robustness are enhanced through the comprehensive application of various machine learning and deep learning methods. Based on the Stacking ensemble learning framework, Extreme Gradient Boosting (XGBoost), K-Nearest Neighbors (KNN), Convolutional Neural Networks (CNN), and Deep Neural Networks (DNN) are used as base learners, each independently analyzing and predicting input features, and their prediction results are then fed as features into a meta-learner. Logistic Regression (LR) is selected as the meta-learner, which learns the relationship between these features and the actual output labels to optimize the final prediction. Through this hierarchical learning process, the Stacking model effectively combines the advantages of different models, improving its ability to identify fault states in the box-flipping robot, thereby enhancing the model's performance and reliability in a production environment.
[0098] In one feasible embodiment, step S3 includes at least the following:
[0099] S31. Acquire real-time data: Collect vibration data of the base platform of the box-flipping robot in real time based on the PLC system, and divide the vibration data into multiple data blocks according to the predetermined working cycle.
[0100] S32. Identify real-time faults based on partitioned data blocks: Input the data blocks generated in each work cycle into the fault identification and detection model to determine the judgment result.
[0101] Specifically, when the fault identification and detection model determines the first label, the operating state is judged as a fault state; when the fault identification and detection model determines the second label, the operating state is judged as a normal state.
[0102] S33. Based on the judgment results, conduct fault warning and response.
[0103] When the fault identification and detection model determines the fault to be the first label, a fault alarm is triggered through the PLC system. Simultaneously, a notification is sent to equipment management personnel via the user interface, clearly indicating the specific timeframe in which the fault occurred, enabling timely troubleshooting and maintenance.
[0104] In summary, the fault identification and detection model for the box-type wire storage and flipping robot provided in this application employs an ensemble learning method, enabling high-precision identification and detection of faulty operating states of the flipping robot. The fault identification and detection model analyzes vibration signal data for each cycle and extracts key time-frequency domain features to construct an input vector. The feature vector is filtered based on a probability density distance method to identify the most representative features. A stacking ensemble learning model is constructed based on these features. This model performs initial predictions by stacking multiple basic learners and optimizes the prediction results using a meta-learner, significantly improving the overall prediction performance of the model. Therefore, the fault identification and detection model provided in this application can monitor the operating status of the flipping robot in real time and accurately warn of potential faults, thereby ensuring production continuity and equipment stability. The prediction accuracy of the fault identification and detection model provided in this application reaches over 95%, enabling timely warnings of equipment faults. Based on the predicted fault location, fault time, and shift information, maintenance personnel can conduct in-depth investigations and repairs in conjunction with business conditions, effectively preventing production interruptions, reducing resource waste, and minimizing unplanned downtime during production.
[0105] By integrating a fault identification and detection model based on ensemble learning into the PLC system, the PLC system collects signal data in real time to monitor the robot's operating status and identify faults.
[0106] By extracting sixteen time-domain statistical features and four frequency-domain features to form vector features, the vibration data of the base platform can be fully utilized.
[0107] Through multi-level feature extraction and filtering, it can focus on overall features while taking into account the matching effects of different features;
[0108] By sorting and dimensionality reduction, the complexity of features is better controlled, thus improving the accuracy of fault identification and detection.
[0109] By integrating multiple machine learning and deep learning algorithms, a complete vibration signal analysis model is constructed for fault identification and detection of box-type wire storage and turning robots.
[0110] By combining the advantages of different algorithms, not only is the accuracy of fault diagnosis improved, but the model's ability to identify complex data patterns is also enhanced.
[0111] By periodically dividing the data, this application can be applied to periodically moving production equipment, providing a basis for improving the efficiency and accuracy of fault identification in periodically moving production equipment and expanding the scope of application of fault identification methods.
[0112] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and substitutions can be made without departing from the technical principles of the present invention, and these improvements and substitutions should also be considered within the scope of protection of the present invention.
Claims
1. A fault identification and detection method for a box-type wire storage and turning robot, characterized in that, Includes the following steps: S1. Obtain vibration data of different operating states of the box-type wire storage and flipping robot, and establish a pre-made database after adding tags to the vibration data; S2. Extract the multi-dimensional feature vector of the vibration data in the prefabricated database, and construct a fault detection and identification model based on the multi-dimensional feature vector; S3. Based on the fault detection and identification model, monitor the operating status of the box-type wire storage and turning robot in real time and issue early warnings.
2. The fault identification and detection method for the box-type wire storage and turning robot according to claim 1, characterized in that, Step S1 includes at least the following: S11. Collect vibration data of the box-type wire storage and flipping robot; the vibration data includes at least vibration data under normal conditions and vibration data under known fault conditions; S12. Preprocessing vibration data: Select vibration data from one working cycle as the processing object, and preprocess the processing object to obtain a pre-made database of vibration data. S13. Add tags to the vibration data in the prefabricated database.
3. The fault identification and detection method for the box-type wire storage and turning robot according to claim 2, characterized in that, In step S11, a vibration sensor is installed at the bottom of the box-type wire storage and flipping robot. The vibration sensor is used to acquire vibration data of the base platform of the box-type wire storage and flipping robot under different operating conditions in real time.
4. The fault identification and detection method for the box-type wire storage and turning robot according to claim 2, characterized in that, In step S13, labels are added to the vibration data in the prefabricated database according to the type of operating state: the data of the fault state is assigned a first label, and the data of the normal operating state is assigned a second label.
5. The fault identification and detection method for the box-type wire storage and turning robot according to claim 1, characterized in that, In step S2, the multidimensional feature vector includes at least a variety of time-domain statistical features and a variety of frequency-domain features.
6. The fault identification and detection method for the box-type wire storage and turning robot according to claim 1, characterized in that, In step S2, the multidimensional feature vectors are sorted based on their importance.
7. The fault identification and detection method for the box-type wire storage and turning robot according to claim 6, characterized in that, Step S2 includes at least the following: S21. Extract the multidimensional feature vectors of vibration data in the prefabricated database, and extract and filter the multidimensional feature vectors; S22. Based on the extracted and filtered feature vectors, a fault identification and detection model is constructed using the stacking ensemble learning algorithm. S23. Train and fit the fault identification and detection model; S24. Determine and store the fault identification and detection model.
8. The fault identification and detection method for the box-type wire storage and turning robot according to claim 1, characterized in that, In step S21, the evaluation index S(d) of multiple feature vectors is obtained sequentially, and the feature vectors are sorted in descending order according to the value of S(d) to complete the selection of feature vectors.
9. The fault identification and detection method for the box-type wire storage and turning robot according to claim 1, characterized in that, Step S23 includes at least the following: S231. Divide the vibration dataset in the prefabricated database into a training set and a test set; S232. Construct a stacking ensemble learning model and initialize relevant model parameters, including the selection of base learners and meta-learners; S233. Train a stacking ensemble learning model using the data from the training set; S234. When the model reaches the iteration completion standard, use the test set to verify the accuracy, and output the predicted labels after completing the evaluation of the model performance.
10. The fault identification and detection method for the box-type wire storage and turning robot according to claim 1, characterized in that, Step S3 includes at least the following: S31. Acquire real-time data: Collect vibration data of the base platform of the box-flipping robot in real time, and divide the vibration data into multiple data blocks according to the predetermined working cycle; S32. Identify real-time faults based on partitioned data blocks: Input the data blocks generated in each work cycle into the fault identification and detection model and determine the output discrimination result; S33. Based on the judgment results, perform fault warning and response.