Prediction method of equipment failure in industrial Internet based on AI big model

Abstract

The present invention relates to the field of equipment failure prediction, and in particular to a method for predicting equipment failures in the industrial Internet based on an AI large model, comprising: obtaining a task action to be learned, and determining an action state according to a change node ratio and an action difference; determining a data processing method according to the action state, the data processing method is to determine a change point selection method according to a data division benchmark and divide the equipment data according to the change points, or to divide the equipment data according to a speed change difference; the data division benchmark is determined according to trajectory complexity; a feature set matching method is determined according to the proportion of affected sections, the feature set matching method is to determine a matching optimization method according to a wear difference coefficient or to match according to a capture influence value, and the matching optimization method is to match according to a wear state or a trajectory reference value to obtain a training feature set. The present invention can improve the accuracy of fault prediction model prediction.

Description

Technical Field

[0001] The present invention relates to the field of equipment failure prediction, and in particular to a method for predicting equipment failures in the industrial Internet based on an AI big model. Background Art

[0002] The Industrial Internet is based on a network system, which ensures the comprehensive interconnection and data flow of people, machines and objects, and can realize the collection, exchange, analysis and optimization of industrial data. At present, robots are common industrial equipment, and intelligent industrial control reduces the use of manpower. The prediction of equipment failure is an important part of industrial control scenarios. In order to reduce the losses caused by equipment failure, the prediction of equipment failure is usually achieved through model learning. However, due to the complexity of the actual motion characteristics of the robot, the accuracy of the equipment failure prediction model is poor. Therefore, how to improve the accuracy of the fault prediction model is a technical problem that needs to be solved urgently by technical personnel in this field.

[0003] Chinese patent publication number CN114495468A discloses a robot fault status monitoring and early warning method, electronic device, and storage medium, the method comprising: S1, collecting input parameters of a fault prediction model; S2, preprocessing the input parameters; S3, using deep learning to establish a fault prediction model that can obtain time series features, and extracting and training the preprocessed input parameters obtained in step S2 through the fault prediction model; S4, the fault prediction model predicts the occurrence of faults based on changes in the robot's signals; S5, making final decisions and executing actions to resolve abnormalities based on the prediction results of the fault prediction model. It can be seen that the above technical solution has the following technical problems: it is impossible to learn the actual task action characteristics of the robot to determine the model learning method for different task actions, resulting in a single model learning method that cannot meet actual usage needs. Summary of the invention

[0004] To this end, the present invention provides a method for predicting equipment failures in the industrial Internet based on an AI big model, so as to overcome the problem in the prior art that it is impossible to learn the actual task action characteristics of the robot to determine the model learning method for different task actions, resulting in a single model learning method being unable to meet actual usage needs.

[0005] To achieve the above objectives, the present invention provides a method for predicting equipment failures in the industrial Internet based on an AI big model, comprising:

[0006] Get the task action to be learned, and determine the action state based on the percentage of changed nodes and the action difference;

[0007] Determine the data processing method according to the action state. The data processing method is to determine the change point selection method according to the data division benchmark and divide the device data according to the change point, or divide the device data according to the speed change difference;

[0008] Among them, the change point selection method is to determine the change point according to the node trajectory or the terminal trajectory;

[0009] The data division criterion is determined according to the trajectory complexity, and the data division criterion is the direction change value or the estimated terminal deviation value;

[0010] The feature set matching method is determined according to the proportion of affected paragraphs. The feature set matching method is to determine the matching optimization method according to the wear difference coefficient or to match according to the capture impact value. The matching optimization method is to match according to the wear state or trajectory reference value to obtain the training feature set.

[0011] Furthermore, the action state is determined according to the percentage of changed nodes and the action difference, and the action state includes:

[0012] A first action state in which the change node ratio is greater than or equal to a preset change node ratio and the action difference is greater than or equal to a preset action difference;

[0013] A second action state in which the change node ratio is less than a preset change node ratio or the action difference is less than a preset action difference.

[0014] Further, determining the data processing method according to the action state;

[0015] In the first action state, the data processing method is to determine the change point selection method according to the data division standard, and divide the device data according to the change point;

[0016] In the second action state, the data processing method is to divide the device data according to the speed change difference;

[0017] Dividing the device data according to the speed change difference includes: detecting the speed change difference corresponding to each reference time point, and dividing the device data by taking the reference time point whose speed change difference is greater than the preset speed change difference as a dividing point;

[0018] The reference time points are time points after the reference sequence of the first time point of the single device data.

[0019] Furthermore, the data division criterion is determined according to trajectory complexity;

[0020] If the trajectory complexity is greater than or equal to the preset trajectory complexity, the data division basis is the direction change value;

[0021] If the trajectory complexity is less than the preset trajectory complexity, the data division basis is the estimated terminal deviation value.

[0022] Further, the change point selection method is determined according to the data division benchmark;

[0023] When the data division basis is the direction change value, the change point selection method is to determine the change point according to the node trajectory;

[0024] When determining the change point according to the node trajectory, a combination analysis is performed on each change node. When a combination analysis is performed on a single change node, the change node is recorded as the target node, and other change nodes that are not recorded in the associated combination, excluding the target node, are recorded as reference nodes;

[0025] Record the set of reference nodes and target nodes whose trajectory difference with the target node is less than the preset trajectory difference as an associated combination, and continue to perform combination analysis on the change nodes not recorded in the associated combination until all the change nodes are recorded in the associated combination;

[0026] Select any change node in each associated combination one by one, and record the time point corresponding to the trajectory point in the moving trajectory corresponding to each change node whose direction change value is greater than the preset direction change value as the change point;

[0027] When the data division basis is the estimated terminal deviation value, the change point selection method is to determine the change point according to the terminal trajectory;

[0028] When determining the change point according to the terminal trajectory, the moving trajectory corresponding to the end of the robot arm is recorded as the terminal trajectory, and the time point corresponding to the trajectory point in the terminal trajectory where the estimated terminal deviation value is greater than the preset estimated terminal deviation value is recorded as the change point.

[0029] Furthermore, the device data is divided according to the change points, where:

[0030] Performing division analysis on each change point in the reference order, when performing division analysis on a single change point, recording the change point as the target change point, and recording the change points that are not recorded as invalid division points, excluding the target change point, as reference change points;

[0031] Record each reference change point whose time reference value to the target change point is less than the preset time reference value as an invalid division point, and continue to perform division analysis on the change points that meet the preset conditions until all change points meet the preset conditions;

[0032] The device data is divided based on the change points that are not recorded as invalid division points;

[0033] The preset condition is that no partition analysis is performed and no invalid partition points are recorded.

[0034] Furthermore, the feature set matching method is determined according to the proportion of the affected paragraphs;

[0035] If the proportion of affected sections is greater than or equal to the preset proportion of affected sections, the feature set matching method is to determine the matching optimization method according to the wear difference coefficient;

[0036] If the proportion of affected paragraphs is less than the preset proportion of affected paragraphs, the feature set matching method is to match according to the captured impact value;

[0037] The affected section is a section where the node speed difference value is greater than or equal to a preset node speed difference value or the speed fluctuation mean value is greater than or equal to a preset speed fluctuation mean value.

[0038] Further, the matching optimization method is determined according to the wear difference coefficient;

[0039] If the wear difference coefficient is greater than or equal to the preset wear difference coefficient, the matching optimization method is to match according to the wear state;

[0040] If the wear difference coefficient is less than the preset wear difference coefficient, the matching optimization method is to match according to the trajectory reference value;

[0041] When matching is performed according to the trajectory reference value, the set of each segment whose trajectory reference value is greater than or equal to the preset trajectory reference value and the set of each segment whose trajectory reference value is less than the preset trajectory reference value are both recorded as the training feature set.

[0042] Furthermore, matching is performed according to the wear state, wherein:

[0043] Recording the set of each segmentation section in the first wear state and the set of each segmentation section in the second wear state as a training feature set;

[0044] The wear state is determined according to the wear node ratio and the maximum wear degree, and the wear state includes:

[0045] A first wear state in which the wear node ratio is greater than or equal to a preset wear node ratio or the maximum wear degree is greater than or equal to a preset maximum wear degree;

[0046] A second wear state in which the wear node ratio is less than a preset wear node ratio and the maximum wear degree is less than a preset maximum wear degree.

[0047] Further, matching is performed based on the crawl impact value, where:

[0048] The set of each segmented paragraph whose grasping influence value is greater than or equal to the preset grasping influence value and the set of each segmented paragraph whose grasping influence value is less than the preset grasping influence value are both recorded as the training feature set.

[0049] Compared with the prior art, the beneficial effect of the present invention lies in that, in the technical scheme of the present invention, the action state is determined according to the proportion of changed nodes and the action difference, and the complexity of the task action to be learned is effectively reflected by the proportion of changed nodes and the action difference, and then different data processing methods are adaptively selected according to the action state, so that the selection of data processing method is more in line with the actual application scenario, avoiding the defect of the prior art that the model learning method cannot be adaptively adjusted for different task actions, resulting in poor prediction effect, thereby improving the training effect of the model and the actual application performance of the model.

[0050] Furthermore, in the present invention, the complexity of the mobile trajectory is effectively reflected by the trajectory complexity, and then different data division criteria are adaptively selected according to the trajectory complexity, so that the selection of the data division criteria is more in line with the actual application scenario, and then the change point selection method is determined according to the data division criteria, so that the selection of the change point selection method is more in line with the actual division scenario of the equipment data, avoiding the analysis defects caused by improper data division, and being able to optimize the data processing effect, conduct targeted training on the equipment data, and thus improve the accuracy of the model fault prediction.

[0051] Furthermore, the present invention effectively reflects the difference in node movement by influencing the proportion of paragraphs, and then adaptively selects different feature set matching methods according to the proportion of influencing paragraphs, so that the selection of feature set matching methods is more in line with the actual application scenario, avoiding the disadvantage of poor model prediction accuracy caused by poor matching effect of divided paragraphs, and making the equipment fault prediction model more stable and accurate.

[0052] Furthermore, in the present invention, the wear difference coefficient is used to effectively reflect the difference in the degree of wear corresponding to the divided sections, and then different matching optimization methods are adaptively selected according to the wear difference coefficient, so that the selection of the matching optimization method is more in line with the actual application scenario, avoiding the defect of poor matching effect of the divided sections due to a single matching method, and can perform targeted training on equipment data of different complexity levels, so that the equipment fault prediction model has higher stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 It is a schematic diagram of a method for predicting equipment failures in the industrial Internet based on an AI big model according to the present invention;

[0054] Figure 2 A flow chart of determining a data processing method according to an action state in the present invention;

[0055] Figure 3 This is a flow chart of the present invention for determining a method for selecting a change point according to a data partitioning benchmark;

[0056] Figure 4The present invention is a flow chart of determining a feature set matching method according to the proportion of affected paragraphs. DETAILED DESCRIPTION

[0057] In order to make the objects and advantages of the present invention more clearly understood, the present invention is further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0058] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the protection scope of the present invention.

[0059] It should be noted that, in the description of the present invention, terms such as "up", "down", "left", "right", "inside" and "outside" indicating directions or positional relationships are based on the directions or positional relationships shown in the drawings. This is merely for the convenience of description and does not indicate or imply that the device or element must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it cannot be understood as a limitation on the present invention.

[0060] In addition, it should be noted that in the description of the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For those skilled in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0061] See also Figures 1 to 4 As shown, the present invention provides a method for predicting equipment failures in the industrial Internet based on an AI big model, comprising:

[0062] Get the task action to be learned, and determine the action state based on the percentage of changed nodes and the action difference;

[0063] Determine the data processing method according to the action state. The data processing method is to determine the change point selection method according to the data division benchmark and divide the device data according to the change point, or divide the device data according to the speed change difference;

[0064] Among them, the change point selection method is to determine the change point according to the node trajectory or the terminal trajectory;

[0065] The data division criterion is determined according to the trajectory complexity, and the data division criterion is the direction change value or the estimated terminal deviation value;

[0066] The feature set matching method is determined according to the proportion of affected paragraphs. The feature set matching method is to determine the matching optimization method according to the wear difference coefficient or to match according to the capture impact value. The matching optimization method is to match according to the wear state or trajectory reference value to obtain the training feature set.

[0067] The application scenario of the present invention is the training of a fault prediction model for robots in the industrial Internet. The task actions to be learned include but are not limited to the robot's mechanical arm grabbing cargo actions, mechanical arm carrying cargo actions, mechanical arm placing cargo actions, and mechanical arm stacking cargo actions; the equipment data includes a number of parameter data corresponding to a single task action to be learned, and the parameter data includes monitoring parameters monitored in real time at each time point during the use time. The monitoring parameters include but are not limited to node position, node speed, node temperature, and node noise. The nodes are the joints in the mechanical arm, wherein the time points are set by the user, and a method for setting the time points is provided, with each 1s being recorded as a time point, i.e., the monitoring parameters are recorded once every 1s.

[0068] In the present invention, several historical records are correspondingly provided, and any historical record records the trajectory complexity, estimated terminal deviation value, affected section ratio, node speed difference value and speed fluctuation mean value in the historical process of at least one equipment fault prediction model training, and each historical record corresponds to a qualified mark, which records whether the prediction accuracy of the equipment fault prediction model meets the user requirements, and the qualified mark can be recorded manually.

[0069] Each training feature set is recorded as a training sample set. Each training feature set is a training sample set, which is used for subsequent model training to obtain a fault prediction model for the task action to be learned. In the present invention, a training feature set corresponds to a partial paragraph of the task action to be learned, so that the training of the fault prediction model is more targeted. This is content that is easy for technical personnel in this field to understand and will not be elaborated on in detail.

[0070] Specifically, the action state is determined according to the proportion of changed nodes and the action difference. The action state includes:

[0071] A first action state in which the change node ratio is greater than or equal to a preset change node ratio and the action difference is greater than or equal to a preset action difference;

[0072] A second action state in which the change node ratio is less than a preset change node ratio or the action difference is less than a preset action difference.

[0073] Among them, for a single task action to be learned, the nodes that move in the robot arm are recorded as changed nodes, and the proportion of changed nodes = the number of changed nodes / the total number of all nodes in the robot arm;

[0074] Action difference = (maximum value among the reference values ​​of the influence points corresponding to each change node - minimum value among the reference values ​​of the influence points corresponding to each change node) / maximum value among the reference values ​​of the influence points corresponding to each change node;

[0075] Obtain the movement trajectory corresponding to each change node. The movement trajectory corresponding to a single change node is the connection line of the node positions corresponding to each time point.

[0076] For a single change node, the node position corresponding to each time point is recorded as a trajectory point, and the vector angle corresponding to each trajectory point after the first trajectory point is detected. The confirmation method of the vector angle corresponding to a single trajectory point is as follows: for a trajectory point, the trajectory point is recorded as the target trajectory point, and the vector angle corresponding to the trajectory point adjacent to the target trajectory point and located before the target trajectory point in the moving order is recorded as the target trajectory point. The vector corresponding to the target trajectory point The angle α between them is recorded as the vector angle, ], the vector corresponding to each trajectory point is tangent to the moving trajectory and the direction of the vector is the same as the moving direction, and the trajectory point with a vector angle greater than the preset vector angle is recorded as an impact point; the value of the preset vector angle can be determined by the user according to the actual application scenario, and a preset vector angle value is provided, and the preset vector angle is 50°; the moving order is the order of reaching each trajectory point; for a single change node, the reference value of the impact point is the total number of impact points in the moving trajectory corresponding to the change node;

[0077] The values ​​of the preset change node ratio and the preset action difference can be determined by the user according to the actual application scenario. The larger the value of the preset change node ratio and the smaller the value of the preset action difference, the greater the user's demand for dividing the device data according to the speed change difference. Provided are values ​​of the preset change node ratio and the preset action difference, the preset change node ratio is 70%, and the preset action difference is 30%.

[0078] Specifically, the data processing method is determined according to the action status;

[0079] In the first action state, the data processing method is to determine the change point selection method according to the data division standard, and divide the device data according to the change point;

[0080] In the second action state, the data processing method is to divide the device data according to the speed change difference;

[0081] Dividing the device data according to the speed change difference includes: detecting the speed change difference corresponding to each reference time point, and dividing the device data by taking the reference time point whose speed change difference is greater than the preset speed change difference as a dividing point;

[0082] The reference time points are time points after the reference sequence of the first time point of the single device data.

[0083] The reference order is the time order from early to late. The speed change difference is confirmed by recording a single reference time point as the target point, recording a time point adjacent to the target point and located before the target point in the reference order as a neighboring point, and the speed change difference corresponding to the target point = (the maximum value of the speed change values ​​corresponding to each change node - the minimum value of the speed change values ​​corresponding to each change node) / the maximum value of the speed change values ​​corresponding to each change node. The speed change value is the absolute value of the difference between the speed corresponding to the target point and the speed corresponding to the neighboring point of a single change node.

[0084] The value of the preset speed change difference can be determined by the user according to the actual application scenario. The higher the accuracy of the user's device data division, the smaller the value of the preset speed change difference. A value of the preset speed change difference is provided, and the preset speed change difference is 30%;

[0085] The reference time points at which the speed change difference is greater than the preset speed change difference are used as division points to divide the device data, wherein the device data is segmented based on each division point to obtain several division sections, a single division section contains several device data sections, the device data section includes several parameter data sections corresponding to a single task action to be learned, and the parameter data section includes monitoring parameters corresponding to several time points.

[0086] Specifically, the data division criterion is determined according to the trajectory complexity;

[0087] If the trajectory complexity is greater than or equal to the preset trajectory complexity, the data division basis is the direction change value;

[0088] If the trajectory complexity is less than the preset trajectory complexity, the data division basis is the estimated terminal deviation value.

[0089] Among them, the trajectory complexity is the standard deviation of the vector angle corresponding to each influencing point. The value of the preset trajectory complexity can be determined by the user according to the actual application scenario. The larger the value of the preset trajectory complexity is, the greater the user's demand for selecting the data division basis as the estimated terminal deviation value is. A preset trajectory complexity value is provided, and the detection data division basis is the historical record of the estimated terminal deviation value. The average value of the trajectory complexity corresponding to the historical records that can meet the user's needs is recorded as the preset trajectory complexity;

[0090] Specifically, the change point selection method is determined according to the data division benchmark;

[0091] When the data division basis is the direction change value, the change point selection method is to determine the change point according to the node trajectory;

[0092] When determining the change point according to the node trajectory, a combination analysis is performed on each change node. When a combination analysis is performed on a single change node, the change node is recorded as the target node, and other change nodes that are not recorded in the associated combination, excluding the target node, are recorded as reference nodes;

[0093] Record the set of reference nodes and target nodes whose trajectory difference with the target node is less than the preset trajectory difference as an associated combination, and continue to perform combination analysis on the change nodes not recorded in the associated combination until all the change nodes are recorded in the associated combination;

[0094] Select any change node in each associated combination one by one, and record the time point corresponding to the trajectory point in the moving trajectory corresponding to each change node whose direction change value is greater than the preset direction change value as the change point;

[0095] When the data division basis is the estimated terminal deviation value, the change point selection method is to determine the change point according to the terminal trajectory;

[0096] When determining the change point according to the terminal trajectory, the moving trajectory corresponding to the end of the robot arm is recorded as the terminal trajectory, and the time point corresponding to the trajectory point in the terminal trajectory where the estimated terminal deviation value is greater than the preset estimated terminal deviation value is recorded as the change point.

[0097] Among them, for the trajectory evaluation values ​​corresponding to the two change nodes, the larger trajectory evaluation value is recorded as A, and the smaller trajectory evaluation value is recorded as B, and the trajectory difference = |AB| / A. For a single change node, the trajectory evaluation value = the influence point reference value × the first coefficient + the influence point distribution coefficient × the second coefficient. The influence point distribution coefficient is the average value of the reference distances corresponding to each influence point. For a single influence point, the influence point is recorded as the target influence point, and the influence points adjacent to the target influence point are recorded as reference influence points. The average value of the distance from the target influence point to each reference influence point is recorded as the reference distance. The distance from the target influence point to the reference influence point is the time interval between the time point corresponding to the target influence point and the time point corresponding to the reference influence point.

[0098] The values ​​of the first coefficient and the second coefficient can be obtained by the user through deep learning convolutional neural network learning for historical records. It can be understood that the present invention reflects the degree of change of the node movement trajectory through the trajectory evaluation value. The user can use deep learning to obtain the influence of the influence point reference value and the influence point distribution coefficient on the degree of change of the node movement trajectory according to the historical records, and then select the values ​​of the first coefficient and the second coefficient accordingly. Provide a value of the first coefficient and the second coefficient, the first coefficient is 0.4, and the second coefficient is 0.6;

[0099] The direction change value is confirmed in the following way: for a single trajectory point, the trajectory point is recorded as the target trajectory point, and the trajectory point adjacent to the target trajectory point and located before the target trajectory point in the moving order is recorded as the adjacent trajectory point. The direction change value corresponding to the target trajectory point = |the vector angle corresponding to the target trajectory point - the vector angle corresponding to the adjacent trajectory point|;

[0100] The moving trajectory corresponding to the end of the robot arm is the line connecting the positions of the end effector at each time point. A continuous cycle monitoring period is provided in the present invention. At the end of each monitoring period, an estimated moving trajectory is determined. The duration of the monitoring period can be set according to the needs of the user. The greater the user's demand for monitoring accuracy, the shorter the duration of the monitoring period. A value of a monitoring period is provided, and the monitoring period is 1 minute; the estimated moving trajectory is the shortest route from the end of the robot arm to the working position; the end of the robot arm is the end that is in direct contact with the goods, and the working position is determined according to the specific task action that the robot arm needs to perform. This is content that is easy for technicians in this field to understand, and will not be described in detail. The estimated end deviation value is the shortest distance from the position of the end effector corresponding to a single time point to the estimated moving trajectory.

[0101] The user can determine the values ​​of the preset trajectory difference, the preset direction change value and the preset estimated terminal deviation value according to the actual application scenario. The higher the accuracy of the user's division of device data, the smaller the values ​​of the preset trajectory difference, the preset direction change value and the preset estimated terminal deviation value. Provided are values ​​of a preset trajectory difference, a preset direction change value and a preset estimated terminal deviation value. The preset trajectory difference is 20%, and the preset direction change value is 30°. The average value of the estimated terminal deviation values ​​corresponding to the historical records that can meet the user's needs is recorded as the preset estimated terminal deviation value.

[0102] Specifically, the device data is divided according to the change points, where:

[0103] Performing division analysis on each change point in the reference order, when performing division analysis on a single change point, recording the change point as the target change point, and recording the change points that are not recorded as invalid division points, excluding the target change point, as reference change points;

[0104] Record each reference change point whose time reference value to the target change point is less than the preset time reference value as an invalid division point, and continue to perform division analysis on the change points that meet the preset conditions until all change points meet the preset conditions;

[0105] The device data is divided based on the change points that are not recorded as invalid division points;

[0106] The preset condition is that no partition analysis is performed and no invalid partition points are recorded.

[0107] The time reference value is the time interval between the target change point and the reference change point. The value of the preset time reference value can be determined by the user according to the actual application scenario. The greater the user's requirement for the length of the divided segment, the greater the value of the preset time reference value. A value of the preset time reference value is provided, and the preset time reference value is 5 minutes.

[0108] The device data is divided based on the change points that are not recorded as invalid division points, wherein the change points that are not recorded as invalid division points are recorded as valid division points, and the device data is segmented based on each valid division point to obtain several division sections.

[0109] Specifically, the feature set matching method is determined according to the proportion of affected paragraphs;

[0110] If the proportion of affected sections is greater than or equal to the preset proportion of affected sections, the feature set matching method is to determine the matching optimization method according to the wear difference coefficient;

[0111] If the proportion of affected paragraphs is less than the preset proportion of affected paragraphs, the feature set matching method is to match according to the captured impact value;

[0112] The affected section is a section where the node speed difference value is greater than or equal to a preset node speed difference value or the speed fluctuation mean value is greater than or equal to a preset speed fluctuation mean value.

[0113] Among them, the data segments after the device data corresponding to the task action to be learned are divided are recorded as divided segments, the proportion of affected segments = the number of affected segments in the divided segments / the total number of divided segments, for a single divided segment, the node speed difference value is the maximum value of the speed difference values ​​corresponding to each time point, for a single time point, the time point is recorded as the target time point, the speed difference corresponding to the target time point = the maximum value of the speeds corresponding to each change node at the target time point - the minimum value of the speeds corresponding to each change node at the target time point; for a single divided segment, the speed fluctuation mean is the average value of the speed fluctuation values ​​corresponding to each change node, and the speed fluctuation value is the standard deviation of the speed of a single change node corresponding to each time point in a single divided segment;

[0114] The values ​​of the preset influence paragraph ratio, the preset node speed difference value and the preset speed fluctuation mean value can be determined by the user according to the actual application scenario. The greater the user's demand for matching according to the captured influence value, the greater the value of the preset influence paragraph ratio. A value of the preset influence paragraph ratio is provided, and the historical records of matching according to the captured influence value are detected. The average value of the influence paragraph ratio corresponding to the historical records that can meet the user's needs is recorded as the preset influence paragraph ratio. The smaller the values ​​of the preset node speed difference value and the preset speed fluctuation mean value, the greater the user's demand for dividing the paragraph as an influence paragraph. A value of the preset node speed difference value and the preset speed fluctuation mean value is provided, and the average value of the node speed difference value corresponding to the historical records that can meet the user's needs is recorded as the preset node speed difference value, and the average value of the speed fluctuation mean corresponding to the historical records that can meet the user's needs is recorded as the preset speed fluctuation mean.

[0115] Specifically, the matching optimization method is determined according to the wear difference coefficient;

[0116] If the wear difference coefficient is greater than or equal to the preset wear difference coefficient, the matching optimization method is to match according to the wear state;

[0117] If the wear difference coefficient is less than the preset wear difference coefficient, the matching optimization method is to match according to the trajectory reference value;

[0118] When matching is performed according to the trajectory reference value, the set of each segment whose trajectory reference value is greater than or equal to the preset trajectory reference value and the set of each segment whose trajectory reference value is less than the preset trajectory reference value are both recorded as the training feature set.

[0119] Among them, wear difference coefficient = maximum value of wear degree corresponding to each change node - minimum value of wear degree corresponding to each change node. For a single change node, wear degree = noise intensity mean × noise coefficient + oil viscosity mean × viscosity coefficient. It can be understood that noise will be generated when the nodes of the robotic arm are worn. At the same time, the friction between the nodes will cause tiny particles on the metal surface to fall off, thereby causing the oil viscosity value of the lubricating oil at the joint of the robotic arm to increase. The noise intensity mean is the average value of the noise intensity monitored at the end of each monitoring cycle within the time range corresponding to a single divided section. The oil viscosity mean is the average value of the oil viscosity value monitored at the end of each monitoring cycle within the time range corresponding to a single divided section. The noise intensity and the oil viscosity value are measured at the change node by a sound sensor and a viscosity sensor respectively. This is easy for technicians in this field to understand and will not be described in detail.

[0120] The values ​​of the noise coefficient and the viscosity coefficient can be obtained by the user through deep learning convolutional neural network learning for historical records. It can be understood that the present invention reflects the wear degree of the change node through the wear degree. The user can use deep learning to obtain the influence of the mean noise intensity and the mean oil viscosity on the wear degree of the node change node according to the historical records, and then select the values ​​of the noise coefficient and the viscosity coefficient accordingly. Provide a value of the noise coefficient and the viscosity coefficient, the noise coefficient is 0.5, and the viscosity coefficient is 0.5;

[0121] The trajectory reference value is the average value of the unit trajectory values ​​corresponding to each change node. The unit trajectory value is confirmed by recording a single change node as the target node, and the unit trajectory value corresponding to the target node = the number of influence points in the moving trajectory corresponding to the divided paragraph / the total number of trajectory points in the moving trajectory corresponding to the divided paragraph;

[0122] The values ​​of the preset wear difference coefficient and the preset trajectory reference value can be determined by the user according to the actual application scenario. The larger the value of the preset wear difference coefficient, the greater the user's demand for matching according to the trajectory reference value. A value of the preset wear difference coefficient is provided, and the historical records of matching according to the trajectory reference value are detected. The average value of the wear difference coefficients corresponding to the historical records that can meet the user's needs is recorded as the preset wear difference coefficient. The larger the value of the preset trajectory reference value, the greater the user's demand for dividing complex segments. A value of the preset trajectory reference value is provided, and the average value of the trajectory reference value that can meet the user's needs is recorded as the preset trajectory reference value.

[0123] Specifically, matching is performed according to the wear state, wherein:

[0124] Recording the set of each segmentation section in the first wear state and the set of each segmentation section in the second wear state as a training feature set;

[0125] The wear state is determined according to the wear node ratio and the maximum wear degree, and the wear state includes:

[0126] A first wear state in which the wear node ratio is greater than or equal to a preset wear node ratio or the maximum wear degree is greater than or equal to a preset maximum wear degree;

[0127] A second wear state in which the wear node ratio is less than a preset wear node ratio and the maximum wear degree is less than a preset maximum wear degree.

[0128] Among them, the wear node is a change node whose wear degree is greater than the preset wear degree. For a single divided paragraph, the wear node ratio = the number of wear nodes / the number of change nodes, and the maximum wear degree is the maximum value of the wear degree corresponding to each change node. The values ​​of the preset wear degree, the preset wear node ratio and the preset maximum wear degree can be determined by the user according to the actual application scenario. The smaller the value of the preset wear degree, the greater the user's need to determine the change node as a wear node. A value of the preset wear degree is provided, and the average value of the wear degree corresponding to the historical records that can meet the user's needs is recorded as the preset wear degree. The larger the values ​​of the preset wear node ratio and the preset maximum wear degree, the greater the user's need to divide the divided paragraphs with large wear degrees. A value of the preset wear node ratio and the preset maximum wear degree is provided, and the historical records of each divided paragraph in the first wear state are detected and recorded as the training feature set. The minimum value of the wear node ratio corresponding to the historical records that can meet the user's needs is recorded as the preset wear node ratio, and the minimum value of the maximum wear degree corresponding to the historical records that can meet the user's needs is recorded as the preset maximum wear degree.

[0129] Specifically, matching is performed based on the crawl impact value, where

[0130] The set of each segmented paragraph whose grasping influence value is greater than or equal to the preset grasping influence value and the set of each segmented paragraph whose grasping influence value is less than the preset grasping influence value are both recorded as the training feature set.

[0131] Among them, the grasping influence value = maximum joint torque / grasping quality value, the grasping quality value is the average weight of the grasped goods at the end of each monitoring period within the time corresponding to a single divided section, the maximum joint torque is the maximum value of the torque value corresponding to each joint at the end of each monitoring period within the time corresponding to a single divided section, and the torque value is monitored by a torque sensor. This is easy for technicians in this field to understand and will not be described in detail;

[0132] The value of the preset grasping influence value can be determined by the user according to the actual application scenario. The larger the value of the preset grasping influence value is, the greater the user's demand for dividing the segment with a large torque influence degree is. A value of the preset grasping influence value is provided, and the set of each segment with a grasping influence value greater than or equal to the preset grasping influence value is detected and recorded as a historical record of the training feature set, and the minimum value of the grasping influence value that can meet the user's needs is recorded as the preset grasping influence value.

[0133] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

[0134] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for predicting equipment failure in the industrial Internet based on an AI big model, characterized in that: include: Get the task action to be learned, and determine the action state based on the percentage of changed nodes and the action difference; Determine the data processing method according to the action state to obtain the segmentation segment, the data processing method is to determine the change point selection method according to the data segmentation benchmark and divide the device data according to the change point, or divide the device data according to the speed change difference; Among them, the change point selection method is to determine the change point according to the node trajectory or the terminal trajectory; The data division criterion is determined according to the trajectory complexity, and the data division criterion is the direction change value or the estimated terminal deviation value; The segmented sections whose node speed difference value is greater than or equal to the preset node speed difference value or whose speed fluctuation mean value is greater than or equal to the preset speed fluctuation mean value are recorded as the influence segment, and the feature set matching method is determined according to the proportion of the influence segment. The feature set matching method is to determine the matching optimization method according to the wear difference coefficient or to match according to the grasping influence value. The matching optimization method is to match according to the wear state or trajectory reference value to obtain the training feature set; Model training is performed on the training feature set to obtain a fault prediction model for the task action to be learned.

2. The method for predicting equipment failure in the industrial Internet based on the AI ​​big model according to claim 1 is characterized in that: The action status is determined based on the percentage of changed nodes and the action difference. The action status includes: A first action state in which the change node ratio is greater than or equal to a preset change node ratio and the action difference is greater than or equal to a preset action difference; A second action state in which the change node ratio is less than a preset change node ratio or the action difference is less than a preset action difference.

3. The method for predicting equipment failure in the industrial Internet based on the AI ​​big model according to claim 2 is characterized in that: Determine the data processing method based on the action status, including: In the first action state, the data processing method is to determine the change point selection method according to the data division standard, and divide the device data according to the change point; In the second action state, the data processing method is to divide the device data according to the speed change difference; Dividing the device data according to the speed change difference includes: detecting the speed change difference corresponding to each reference time point, and dividing the device data by taking the reference time point whose speed change difference is greater than the preset speed change difference as a dividing point; The reference time points are time points after the reference sequence of the first time point of the single device data.

4. The method for predicting equipment failure in the industrial Internet based on the AI ​​big model according to claim 3 is characterized in that: The data division criterion is determined according to the trajectory complexity, including: If the trajectory complexity is greater than or equal to the preset trajectory complexity, the data division basis is the direction change value; If the trajectory complexity is less than the preset trajectory complexity, the data division basis is the estimated terminal deviation value.

5. The method for predicting equipment failure in the industrial Internet based on the AI ​​big model according to claim 4 is characterized in that: Determine the change point selection method based on the data division benchmark, including: When the data division basis is the direction change value, the change point selection method is to determine the change point according to the node trajectory; When determining the change point according to the node trajectory, a combination analysis is performed on each change node. When a combination analysis is performed on a single change node, the change node is recorded as the target node, and other change nodes that are not recorded in the associated combination, excluding the target node, are recorded as reference nodes; Record the set of reference nodes and target nodes whose trajectory difference with the target node is less than the preset trajectory difference as an associated combination, and continue to perform combination analysis on the change nodes not recorded in the associated combination until all the change nodes are recorded in the associated combination; Select any change node in each associated combination one by one, and record the time point corresponding to the trajectory point in the moving trajectory corresponding to each change node whose direction change value is greater than the preset direction change value as the change point; When the data division basis is the estimated terminal deviation value, the change point selection method is to determine the change point according to the terminal trajectory; When determining the change point according to the terminal trajectory, the moving trajectory corresponding to the end of the robot arm is recorded as the terminal trajectory, and the time point corresponding to the trajectory point in the terminal trajectory where the estimated terminal deviation value is greater than the preset estimated terminal deviation value is recorded as the change point.

6. The method for predicting equipment failure in the industrial Internet based on the AI ​​big model according to claim 5 is characterized in that: The device data is divided according to the change points, among which, Performing division analysis on each change point in the reference order, when performing division analysis on a single change point, recording the change point as the target change point, and recording the change points that are not recorded as invalid division points, excluding the target change point, as reference change points; Record each reference change point whose time reference value to the target change point is less than the preset time reference value as an invalid division point, and continue to perform division analysis on the change points that meet the preset conditions until all change points meet the preset conditions; The device data is divided based on the change points that are not recorded as invalid division points; The preset condition is that no partition analysis is performed and no invalid partition points are recorded.

7. The method for predicting equipment failure in the industrial Internet based on the AI ​​big model according to claim 6 is characterized in that: Determine the feature set matching method based on the proportion of affected paragraphs, including: If the proportion of affected sections is greater than or equal to the preset proportion of affected sections, the feature set matching method is to determine the matching optimization method according to the wear difference coefficient; If the affected paragraph ratio is less than the preset affected paragraph ratio, the feature set matching method is to match according to the captured impact value.

8. The method for predicting equipment failure in the industrial Internet based on the AI ​​big model according to claim 7 is characterized in that: Determine the matching optimization method based on the wear difference coefficient, including: If the wear difference coefficient is greater than or equal to the preset wear difference coefficient, the matching optimization method is to match according to the wear state; If the wear difference coefficient is less than the preset wear difference coefficient, the matching optimization method is to match according to the trajectory reference value; When matching is performed according to the trajectory reference value, the set of each segment whose trajectory reference value is greater than or equal to the preset trajectory reference value and the set of each segment whose trajectory reference value is less than the preset trajectory reference value are both recorded as the training feature set.

9. The method for predicting equipment failure in the industrial Internet based on the AI ​​big model according to claim 8 is characterized in that: Matching is performed according to the wear status, where Recording the set of each segmentation section in the first wear state and the set of each segmentation section in the second wear state as a training feature set; The wear state is determined according to the wear node ratio and the maximum wear degree, and the wear state includes: A first wear state in which the wear node ratio is greater than or equal to a preset wear node ratio or the maximum wear degree is greater than or equal to a preset maximum wear degree; A second wear state in which the wear node ratio is less than a preset wear node ratio and the maximum wear degree is less than a preset maximum wear degree.

10. The method for predicting equipment failure in the industrial Internet based on the AI ​​big model according to claim 7 is characterized in that: Matching is done based on the crawl impact value, where The set of each segmented paragraph whose grasping influence value is greater than or equal to the preset grasping influence value and the set of each segmented paragraph whose grasping influence value is less than the preset grasping influence value are both recorded as the training feature set.

Images