Bolt robot unscrewing method based on machine learning

Abstract

The invention discloses a bolt robot screwing-out method based on machine learning, and the method comprises the steps: S1, collecting a bolt disassembly force data sample and a corresponding bolt disassembly speed, extracting key force data features in a specific mode, and constructing a sample data set; s2, training the distribution of the selected different time series data prediction models by adopting the sample data set, and selecting an optimal data prediction model based on model prediction performance; s3, the collected bolt disassembling force data are input into the optimal data prediction model, and the upward moving speed of the robot is obtained; and S4, the upward moving speed of the robot is set into a robot program, and the robot flexibly disassembles the bolt. According to the method, the disassembly efficiency of the bolts can be improved, a more accurate and flexible disassembly scheme is provided for the field of automatic disassembly of the bolts, the problem of interference between the robot and the bolts during bolt disassembly is effectively solved, and the adaptability of the robot for disassembling the bolts is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of robot bolt disassembling, and particularly to a method for a bolt robot to unscrew based on machine learning. Background Art

[0002] Remanufacturing is a process of processing and refurbishing retired machinery and equipment into products with the same performance as new ones. In the remanufacturing process, disassembly is the key first step. Systematically disassembling retired machinery and equipment can improve the recovery rate of retired machinery and equipment and promote the economic benefits of enterprises. At present, the disassembly of retired machinery and equipment is mainly completed manually. However, the manual disassembly method has high labor intensity and low efficiency, and there are the following problems: Manual disassembly requires operators to maintain a high-intensity working state for a long time, which is prone to fatigue and occupational diseases; The manual disassembly speed is slow and it is difficult to meet the needs of large-scale disassembly of retired equipment; It is difficult to ensure the accuracy of each operation during manual disassembly, and it is easy to cause equipment damage or part loss; There are certain safety risks during manual disassembly, such as tool slippage and part splashing. An automated disassembly method is needed to solve this problem.

[0003] Robots have become one of the main methods of automated disassembly because of their flexibility and interactivity. Bolt connection is the main connection method of retired machinery and equipment. According to research statistics, about 40% of the operations in the disassembly process of retired machinery and equipment are related to bolt disassembly. Therefore, realizing automated bolt disassembly by robots is a necessary step to achieve automated disassembly of retired mechanical equipment.

[0004] Therefore, it is an urgent problem for those skilled in the art to propose a method for a bolt robot to unscrew based on machine learning to solve the difficulties existing in the prior art. Summary of the Invention

[0005] The purpose of the present invention is to provide a method for a bolt robot to unscrew based on machine learning, which not only helps to improve the disassembly efficiency of bolts, but also provides a more accurate and flexible disassembly solution for the field of automated bolt disassembly.

[0006] To achieve the above purpose, the present invention provides the following solution:

[0007] A method for a bolt robot to unscrew based on machine learning includes the following steps:

[0008] S1. Collect bolt disassembly force data samples and the corresponding bolt disassembly speeds, extract key force data features under specific modes, and construct a sample data set;

[0009] S2. Use the sample data set to train the selected different time series data prediction model distributions, and select the optimal data prediction model based on the model prediction performance.

[0010] S3. Input the collected bolt disassembly force data into the optimal data prediction model to obtain the upward movement speed of the robot.

[0011] S4. Set the upward movement speed of the robot in the robot program to achieve flexible disassembly of bolts by the robot.

[0012] Preferably, in S1, the bolt disassembly force data specifically is: the robot rigid bolt disassembly force data along the main axis direction collected by the sensor based on the selected bolt model and the preset rotating speed of the disassembler.

[0013] Preferably, in S1, extracting the key force data features in a specific mode specifically includes:

[0014] Delete the part of the bolt disassembly force data that exceeds the preset threshold; fill the preset value after the bolt disassembly force data sequence until the preset sequence length, and the preset value is the minimum value of the bolt disassembly force data after the deletion process; filter out the waveforms higher than the specified Hertz in the bolt disassembly force data after the deletion and filling process; perform data normalization on the bolt disassembly force data after the deletion, filling, and filtering process; divide the sample data set, and the sample data set is divided into a training set and a test set.

[0015] Preferably, in S2, use the same data set to train the selected different time series data prediction model distributions. The different time series data prediction models include: one-dimensional convolutional neural network 1D CNN, multi-layer perceptron MLP, long short-term memory neural network LSTM, convolutional neural network-long short-term memory neural network CNN-LSTM.

[0016] Preferably, in S2, select the optimal data prediction model based on the model prediction performance. Among them, the model prediction performance includes: the proportion of the number of correctly predicted labels in the prediction results obtained by inputting the test set into the optimal data prediction model to the total number of the test set.

[0017] Preferably, in S4, the robot flexibly disassembling the bolt specifically is:

[0018] Based on the selected bolt model and the preset rotating speed of the disassembler, after the robot moves upward with the speed parameter obtained by the optimal data prediction model, the rising speed of the robot is consistent with that of the bolt, avoiding rigid collision.

[0019] Preferably, the process of screwing out the bolt is divided into three stages: a linear stage, a stable stage, and a disassembly termination judgment stage:

[0020] After the disassembly system is started, it first enters the linear stage. The operating end robot remains stationary, the bolt driver drives the sleeve to rotate, and the bolt moves upward while generating a linearly increasing contact force with the robot; the control end collects the original contact force data in the linear stage. When the contact force increases to the specified threshold, it enters the next stage;

[0021] In the steady stage, the control end preprocesses the original contact force data in the linear stage and inputs it into the trained optimal data prediction model to obtain the optimal rising speed of the robot. The robot at the operation end drives the bolt driver to rise, the contact force tends to be stable, and the unscrewing operation proceeds smoothly until the next stage; entering the termination signal determination stage, when the bolt unscrewing is completed, the robot keeps moving up, releases the contact force, stops rotating, lifts the bolt to the specified position, and the disassembly is completed.

[0022] Preferably, when the bolt unscrewing process is in the linear stage, the contact force F in the linear stage z and the bolt unscrewing speed V s have the following relationship:

[0023] F z = kv s t + b z

[0024] where k is a constant coefficient, b z is the bias, and t is the time.

[0025] The present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements a bolt robot unscrewing method based on machine learning as described in any one of the above.

[0026] According to the specific embodiments provided by the present invention, the following technical effects are disclosed:

[0027] (1) For the bolt robot unscrewing method based on machine learning of the present invention, the robot disassembly system built for bolt automatic disassembly integrates sensors, bolt drivers, and depth camera hardware, and integrates the communication control programs of each hardware onto the same upper computer, which can collect six-dimensional force / torque data during the bolt disassembly process in real time and execute the disassembly commands obtained from algorithm analysis. The entire disassembly system can make good use of the application points where machine learning neural network algorithms can be deployed on the upper computer, improving the disassembly accuracy rate and automation level of the disassembly plan.

[0028] (2) For the bolt robot unscrewing method based on machine learning of the present invention, the bolt disassembly force data collected at the beginning is input into the trained convolutional neural network model to obtain the rising speed of the robot; the robot rising speed parameter is set into the robot program to realize the flexible disassembly of the bolt by the robot; this method not only helps to improve the disassembly efficiency of the bolt, but also provides a more accurate and flexible disassembly plan for the field of bolt automatic disassembly. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the accompanying drawings required in the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can also be obtained based on these drawings.

[0030] Figure 1 It is a step diagram of a robot disassembling a bolt using the bolt robot unscrewing method based on machine learning of the present invention;

[0031] Figure 2 It is a schematic diagram of the relative distance between a bolt and a sleeve when the bolt is unscrewed in an embodiment of the present invention;

[0032] Figure 3 It is the force time series data of the main axis direction of a bolt driver collected by a sensor when a robot unscrews a bolt in an embodiment of the present invention;

[0033] Figure 4 It is a schematic diagram of a robot being similar to a spring when unscrewing a bolt in an embodiment of the present invention;

[0034] Figure 5 It is a result diagram of training a model using different machine learning algorithms and the same data set in an embodiment of the present invention. Detailed implementation manners

[0035] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0036] To make the above objects, features, and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific implementation manners.

[0037] A bolt robot unscrewing method based on machine learning provided by the present invention includes the following steps:

[0038] S1. Collect bolt disassembly force data samples and corresponding bolt disassembly speeds, extract key force data features in a specific mode, and construct a sample data set;

[0039] S2. Use the sample data set to train different time series data prediction model distributions selected, and select the optimal data prediction model based on the model prediction performance;

[0040] S3. Input the collected bolt disassembly force data into the optimal data prediction model to obtain the upward movement speed of the robot.

[0041] S4. Set the upward movement speed of the robot in the robot program to achieve flexible disassembly of bolts by the robot.

[0042] See Figure 1 As shown, it is the disassembly step diagram of the embodiment of the present invention. The process of unscrewing the bolt is divided into three stages: the linear stage, the stable stage, and the disassembly termination judgment stage. After the disassembly system is started, it first enters the linear stage. In this stage, the operating-end robot remains stationary, and the bolt driver drives the sleeve to rotate. The bolt moves upward and generates a linearly increasing contact force F with the robot z ; the control end collects the original data of the contact force F in the linear stage z . When F z increases to the specified threshold, it enters the next stage. In the stable stage, the control end preprocesses the original data of the contact force in the linear stage and inputs it into the trained machine learning model to obtain the optimal upward movement speed V of the robot r . The operating-end robot drives the bolt driver to rise, the contact force tends to be stable, and the unscrewing operation proceeds smoothly until the next stage. Enter the termination signal determination stage. When the bolt unscrewing is completed, the robot keeps moving upward and releases the contact force. F z increases or decreases to 0 N. Based on this feature, stop rotating, lift the bolt to the specified position, and the disassembly is completed.

[0043] In the method for unscrewing bolts by a bolt robot based on machine learning provided by the embodiment of the present invention, the relationship model between the force in the main axis direction of the bolt driver and the bolt detachment plane speed during the bolt disassembly process includes: during the bolt unscrewing process, the bolt unscrewing speed can be expressed as:

[0044] V s = npz

[0045] In the formula: V s is the speed at which the bolt unscrews along the main axis direction; n is the rotation speed of the bolt disassembler; p is the pitch of the bolt; z is the number of bolt heads.

[0046] When the upward movement speed V of the robot r < V s , the sleeve will interfere with the bolt, and the contact force F z will gradually increase. When the upward movement speed V of the robot r > V s , the sleeve and the bolt will move away from each other until the sleeve detaches from the bolt. Only when V r = V s , dz remains unchanged, and the bolt unscrewing state is stable as Figure 2 shown.

[0047] Set V r When it is 0, the force in the F z direction and the image of force versus time show a linear regression relationship as Figure 3 shown. When the robot is stationary, the bolt rotates upward and interferes with the robot to generate a contact force. This interference distance will be borne by the elasticity of the robot's joints. At this time, the robot is like a spring as Figure 4 shown. Therefore, when the robot is stationary and the disassembler rotates at a constant speed, the interference distance increases with time, and the contact force also increases with the interference distance. The image shows a linear regression relationship until the contact force rises to the force limit protection of the robot or the bolt disassembly is completed, and the image tends to be flat.

[0048] The contact force Fz can be expressed as:

[0049] F z = kh + b z

[0050] That is, the contact force F z and the compression displacement h in the main axis direction have a linear relationship, k is a constant coefficient, and b z is the offset. At the same time, h can be expressed as:

[0051] dh = (v s - v r )dt

[0052] h = ∑dh

[0053] When n and V r are fixed values, there is F z The formula is as follows:

[0054] F z = k(v s - v r )t + b z

[0055] When the screwing-out process is in the linear stage, V r = 0. The relationship between the contact force F z and the bolt screwing-out speed V s is as follows:

[0056] F z = kv s t + b z

[0057] At this time, V s is the optimal robot rising speed, and V s represents among the slopes of the contact force F z versus time t.

[0058] When the bolt information is known, V sIt can be calculated by a formula, but the parameter information of the actual disassembled bolt is unknown. At the same time, there are sensor noises and system disturbances in the contact force data, which cannot fully conform to the ideal linear state. Therefore, the automatic disassembly of the bolt requires an intelligent algorithm model to establish the relationship between the contact force data and the optimal rising speed of the robot, so as to achieve accurate stopping.

[0059] Furthermore, in S1, the bolt disassembly force data specifically is: the robot rigid bolt disassembly force data along the main axis direction collected by the sensor based on the selected bolt model and the preset rotation speed of the disassembler.

[0060] Specifically, the characteristics of the sample data set in S1 are the forces in the main axis direction of the bolt driver collected by the sensor when the robot disassembly system uses bolts of models M4, M6, and M8, and the bolt driver disassembles at 100 and 200 r / mmin respectively; the labels are 70, 100, 125, 140, 200, 250 mm / mmin obtained by substituting the bolt pitch and the bolt driver rotation speed into the formula; the training set and the test set are divided into 8:2 according to the labels.

[0061] Even further, in S1, extract the key data characteristics under specific modes, including:

[0062] Delete the part of the bolt disassembly force data that exceeds the preset threshold; fill in the preset value after the bolt disassembly force data sequence until the preset sequence length, and the preset value is the minimum value of the bolt disassembly force data after the deletion process; filter out the waveforms higher than the specified hertz in the bolt disassembly force data after the deletion and filling process, and perform data normalization on the bolt disassembly force data after the deletion, filling, and filtering process; divide the sample data set, and the sample data set is divided into a training set and a test set.

[0063] Specifically, intercept the force data in the range of 5 - 40 N as the model feature; take 70 data as the unified input sequence length, truncate the part of the data that is more than 70 in the direction from 40 N to 5 N, and fill the part with insufficient 70 sequence length with the minimum value of the interval data; perform Max - Min normalization and low - pass filtering on the data.

[0064] Specifically, in S2, the same data set is used to train the prediction model distributions of different time - series data. The different time - series data prediction models include five machine learning models: one - dimensional convolutional neural network 1D CNN, multi - layer perceptron MLP, long short - term memory neural network LSTM, and convolutional neural network - long short - term memory neural network CNN - LSTM. The input of the model is the time - series data of the force in the main axis direction of the bolt driver, and the output is the optimal lifting speed of the robot.

[0065] Specifically, in S2, the optimal machine learning method is selected based on the prediction performance, where the prediction performance includes: the proportion of the number of correctly predicted labels in the prediction results obtained by inputting the test set features into the model to the total number of the test set. The closer the proportion is to 1.0, the better the performance; the better the performance of the training model in achieving a higher accuracy rate within a shorter period. Refer to Figure 5 As shown, it is the result graph of training the model using different machine learning algorithms and the same data set in the embodiment of the present invention. The CNN has the best prediction performance, with an accuracy rate reaching 96%, and at the same time achieving a higher accuracy rate within a shorter period.

[0066] Specifically, the upward movement speed of the robot in S3 refers to the speed at which the bolt detaches from the disassembly plane obtained by inputting the force time series data in the direction of the main shaft of the bolt driver at the beginning part into the convolutional neural network model under the specified bolt model and the rotation speed of the bolt driver.

[0067] Specifically, the flexible disassembly of the bolt by the robot in S4 means that under the specified bolt model and the rotation speed of the bolt driver, after the robot is lifted according to the speed parameter obtained by the convolutional neural network, since the lifting speed of the robot is consistent with the speed at which the bolt detaches from the plane, it avoids the rigid collision between the sleeve and the bolt when the lifting speed of the robot is less than the speed at which the bolt detaches from the plane and the mating failure between the sleeve and the bolt when the lifting speed of the robot is greater than the speed at which the bolt detaches from the plane during the disassembly process.

[0068] The rigid disassembly of the bolt by the robot is specifically: the disassembly method in which the robot keeps the initial disassembly position unchanged and the disassembler drives the bolt to rotate counterclockwise.

[0069] The present invention also provides a non - transient computer - readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements a bolt robot unscrewing method based on machine learning as described in any one of the above.

[0070] It also includes a bolt robot unscrewing system based on machine learning, including: Intel realsense D435 series cameras, UR5 robots, Robotiq FT300 force sensors, and Robotiq SD100 bolt drivers. The Robotiq SD100 bolt driver and the Robotiq FT300 force sensor are sequentially connected to the end effector of the UR5 robot and connected to the control cabinet of the UR5 robot, and then integrated with the upper computer using an Ethernet interface and interact using the TCP / IP protocol.

[0071] The Intel RealSense D435 series cameras are integrated with the host computer through the USB interface and interact using the RealSense SDK 2.0. In terms of specific functions, the UR5 robot and the Robotiq SD100 bolt driver form the operating end to execute the disassembly command; the Robotiq FT300 force sensor and the host computer form the control end to process the input information and issue instructions.

[0072] A method for unscrewing bolts of a bolt robot based on machine learning constructs a relationship model between the force in the main axis direction of the bolt driver and the bolt detachment plane speed during the bolt disassembly process, explaining the principle that the force time series data in the main axis direction of the bolt driver contains bolt detachment plane speed information. The above relationship model can help researchers analyze the research value in the disassembly force time series data and further develop an intelligent disassembly plan in combination with machine learning algorithms.

[0073] This method uses the convolutional neural network algorithm to establish the mapping relationship between the disassembly force time series data with system noise and the optimal lifting speed during the robot disassembly process. This algorithm can identify the characteristics of the disassembly force time series data for different disassembly speeds and bolt models, improving the disassembly success rate of the bolt disassembly plan.

[0074] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus the necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution, in essence, or the part that contributes to the prior art can be embodied in the form of a software product, which can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0075] Specific examples are used in this article to elaborate on the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method and its core idea of the present invention; at the same time, for those of ordinary skill in the art, based on the idea of the present invention, there will be changes in the specific implementation methods and application scopes. In summary, the content of this specification should not be construed as a limitation to the present invention.

Claims

1. A bolt unscrewing method based on machine learning, characterized in that: The following steps are involved: S1. Collect bolt disassembly force data samples and corresponding bolt disassembly speeds, extract key force data features in specific modes, and construct a sample data set; S2. Using the sample data set to train the selected different time series data prediction model distributions, and selecting the optimal data prediction model based on the model prediction performance; S3, inputting the collected bolt disassembly force data into the optimal data prediction model to obtain the robot upward movement speed; S4. Set the robot upward speed in the robot program to enable the robot to flexibly disassemble the bolts.

2. The bolt unscrewing method based on machine learning according to claim 1, characterized in that: In S1, the bolt disassembly force data specifically includes: robot rigid bolt disassembly force data along the main axis direction collected by the sensor based on the selected bolt model and the preset disassembler rotation speed.

3. The bolt unscrewing method based on machine learning according to claim 1, characterized in that: In S1, extracting key force data features in a specific mode specifically includes: Delete part of the bolt disassembly force data that exceeds a preset threshold; fill a preset value after the bolt disassembly force data sequence to a preset sequence length, wherein the preset value is the minimum value of the bolt disassembly force data after the deletion process; filter out waveforms higher than a specified Hz in the bolt disassembly force data after the deletion and filling process; perform data normalization on the bolt disassembly force data after the deletion and filling filtering process; and divide the sample data set into a training set and a test set.

4. The bolt unscrewing method based on machine learning according to claim 1, characterized in that: In S2, the same data set is used to train the selected different time series data prediction model distributions, and the different time series data prediction models include: one-dimensional convolutional neural network 1D CNN, multi-layer perceptron MLP, long short-term memory neural network LSTM, convolutional neural network-long short-term memory neural network CNN-LSTM.

5. The bolt unscrewing method based on machine learning by a robot according to claim 4, characterized in that: In S2, the optimal data prediction model is selected based on the model prediction performance, wherein the model prediction performance includes: the ratio of the number of correct label predictions in the prediction results obtained by the optimal data prediction model when the test set is input to the test set to the number of correct label predictions in the entire test set.

6. The bolt unscrewing method based on machine learning by a robot according to claim 1, characterized in that: In S4, the robot flexibly dismantles the bolts as follows: Based on the selected bolt model and the preset disassembler rotation speed, after the robot moves up according to the speed parameters obtained by the optimal data prediction model, the robot and the bolt rise at the same speed to avoid rigid collision.

7. The bolt unscrewing method based on machine learning according to claim 1, characterized in that: The bolt unscrewing process is divided into three stages: linear stage, stable stage and disassembly termination judgment stage: After the disassembly system is started, it first enters the linear stage. The robot at the operating end remains stationary, and the bolt driver drives the sleeve to rotate. The bolt moves upward and generates a linearly increasing contact force with the robot. The control end collects the original data of the contact force in the linear stage, and enters the next stage when the contact force increases to the specified threshold. In the stable stage, the control end pre-processes the original contact force data of the linear stage and inputs it into the trained optimal data prediction model to obtain the optimal robot rising speed. The operating end robot drives the bolt driver to rise, the contact force tends to be stable, and the unscrewing operation proceeds smoothly until the next stage; entering the termination signal judgment stage, when the bolt is unscrewed, the robot keeps moving up, releases the contact force, stops rotating, lifts the bolt to the specified position, and the disassembly is completed.

8. The bolt unscrewing method based on machine learning by a robot according to claim 7, characterized in that: When the bolt unscrewing process is in the linear stage, the linear stage contact force F z and bolt rotation speed V s The relationship is as follows: F z =kv s t+b z Among them, k is a constant coefficient, bz is a bias, and t is time.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the bolt unscrewing method based on machine learning by a robot is implemented as described in any one of claims 1 to 8.

Images