Method and system for predicting equipment health state based on large model AGENT capability

Abstract

The invention relates to the technical field of numerical control machine tools, and discloses a method and system for predicting the health state of equipment based on the large model AGENT capacity, and the method comprises the following steps: obtaining the machining data of a numerical control machine tool; constructing plane data based on the machining data of the numerical control machine tool; inputting the plane data into a decision model, and outputting a machining scheme of the numerical control machine tool and a risk level of the machining scheme by the decision model; finely adjusting parameters of the decision model by using transfer learning, wherein the finely-adjusted decision model is used for an actual numerical control machine tool machining process; according to the system, machining data of the numerical control machine tool are collected regularly, machining experience and knowledge of the existing numerical control machine tool are migrated to a new machining task, and compared with the mode that the needed machining effect is achieved through a large number of experiments and parameter adjustment, the system can improve the machining efficiency under the condition that safety is ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of numerically controlled machine tools, and more specifically, to a method and system for predicting the health status of equipment based on large model AGENT capability. Background Art

[0002] With the rapid development of modern manufacturing, CNC machine tools have been widely used in various industries. The use of deep learning technology can transfer the processing experience and knowledge of existing CNC machine tools to new processing tasks, which can significantly improve the processing efficiency and quality. However, the CNC machine tool processing process involves high-speed motion and cutting force. If the generated processing plan has defects, it may lead to the scrapping of workpieces, damage to equipment and even personal safety accidents, resulting in high economic losses. Summary of the invention

[0003] The purpose of the present invention is to provide a method and system for predicting the health status of equipment based on the ability of a large model AGENT in order to solve the above problems.

[0004] The present invention provides a method for predicting the health status of equipment based on the large model AGENT capability, comprising the following steps:

[0005] Step 100, acquiring the processing data of the CNC machine tool. The processing data of the CNC machine tool is collected once at each time sampling point, including the processing data of the spindle system, the feeding system, the tool magazine and tool changing system, the cooling and lubrication system, the workpiece clamping and positioning system, the on-machine measurement system and the processing task data of the CNC machine tool;

[0006] Step 200, constructing plane data based on the processing data of the numerical control machine tool, including facets for storing data, a first serial data stream of a facet indexing a data object, and the plane data further including a relationship list for storing association relationships between data objects indexed by the facets, wherein the element value of the i-th row and j-th column of the relationship list represents a quantitative value of the association relationship between the data objects indexed by the i-th and j-th facets;

[0007] The association relationships include high and low, corresponding to two quantitative values;

[0008] Data objects include spindle system, feed system, tool magazine and tool changing system, cooling and lubrication system, workpiece clamping and positioning system, on-machine measurement system, and processing tasks;

[0009] A serial data stream No. 1 of a data object includes n data units ordered by time, and the t-th data unit represents the processing data of the CNC machine tool collected at the t-th moment of the data object;

[0010] Step 300, inputting the plane data into a decision model, the decision model comprising a backbone network, the backbone network comprising a first intermediate layer, a second intermediate layer, a third intermediate layer, a fourth intermediate layer, a fifth intermediate layer, a first FC layer and a second FC layer, the first FC layer outputting a processing plan of a CNC machine tool, and the second FC layer outputting a risk level of the processing plan;

[0011] Step 400, using transfer learning to fine-tune the parameters of the decision model, and the fine-tuned decision model is used in the actual CNC machine tool processing process.

[0012] Furthermore, the time sampling point refers to the moment at which data is collected at a predetermined time interval, and the default value of the predetermined time interval is 1 minute.

[0013] Further, the spindle system processing data includes spindle speed, spindle torque, spindle power, spindle temperature and spindle vibration data;

[0014] The feed system processing data includes feed speed, cutting depth, cutting width, X, Y, Z axis coordinates and A, B, C axis angle data;

[0015] The processing data of tool magazine and tool changing system include tool type, tool material, tool geometry parameters, tool wear status and tool changing time data;

[0016] The processing data of cooling and lubrication system include coolant temperature, coolant pressure, coolant flow, lubricating oil temperature and lubricating oil pressure data;

[0017] The workpiece clamping and positioning system processing data includes chuck pressure, jaw position, workpiece positioning accuracy and workpiece clamping force data;

[0018] The processing data of the on-machine measurement system includes the probe trigger signal, probe position coordinates, probe pressure and measurement result data;

[0019] The processing task data of CNC machine tools include material properties, processing technology, processing procedures, workpiece drawings, processing time, and processing quality requirements.

[0020] Furthermore, data objects with high association are defined as follows: there is a relationship between the spindle system and the feed system, cooling and lubrication system, tool magazine and tool changing system; there is a relationship between the feed system and the workpiece clamping and positioning system, and the on-machine measurement system; there is a relationship between the processing task and all other data objects.

[0021] Further, the first serial data stream is input into the first intermediate layer, the first identification feature is output to the second intermediate layer, the second intermediate layer also inputs the relationship list, and the second intermediate layer outputs the second identification feature; the first serial data stream is input into the third intermediate layer, the third identification feature is output to the fourth intermediate layer, the fourth intermediate layer also inputs the relationship list, the fourth intermediate layer outputs the fourth identification feature to the second feature fusion layer, and the second feature fusion layer outputs the sixth fusion state to the first feature fusion layer; the first feature fusion layer inputs the second identification feature and the sixth fusion state, and outputs the seventh fusion state to the first FC layer; the second FC layer inputs the sixth fusion state and outputs the risk level;

[0022] In the first running step, the second feature fusion layer inputs the fourth recognition feature and a default state in which both components are zero values, and the size of the default state is the same as the fifth hidden state. The first FC layer outputs the initial processing plan, which includes the processing parameters of the spindle system, the feed system, the tool magazine and the tool changing system, the cooling and lubrication system, the workpiece clamping and positioning system, and the on-machine measurement system, including the spindle speed setting value, the spindle torque limit value, the spindle power limit value, the spindle temperature alarm value, the feed speed setting value, the cutting depth setting value, the cutting width setting value, the motion instructions of the X, Y, Z axis coordinates and the A, B, C axis angles, the tool selection instructions, the tool changing instructions, the tool compensation parameters, the coolant temperature setting value, the coolant pressure setting value, the coolant flow setting value, the lubricating oil temperature setting value, the lubricating oil pressure setting value, the chuck pressure setting value, the jaw position setting value, the workpiece positioning coordinate setting value, the workpiece clamping force setting value, the probe selection instruction, the measurement instruction, and the measurement coordinate setting value;

[0023] In the second operation step, the processing plan outputted by the first FC layer for the last time is used to generate a second serial data stream, and a data unit of the second serial data stream represents a processing parameter at a moment in the processing plan outputted by the first FC layer for the last time; the second serial data stream is inputted into the fifth intermediate layer, and the fifth recognition feature is outputted to the second feature fusion layer; the first FC layer outputs the intermediate processing plan, and if the risk level outputted by the second FC layer is low risk, the initial processing plan is outputted as the final processing plan; if the risk level is medium risk, the intermediate processing plan is outputted as the final processing plan; if the risk level is high risk, the second operation step is run again.

[0024] Furthermore, the formula for the first intermediate layer is as follows:

[0025] ;

[0026] ;

[0027] ;

[0028] in Represents the plane data The first identification feature of the t-th data unit of the serial data stream No. 1 corresponding to the facet, Indicates The t-1th first recognition feature of the face element, and Respectively represent The first and second intermediate features of the facet, Indicates The tth data unit of the serial data stream No. 1 of bins, and are the first and second serial recognition weight matrices, is the first serial identification bias parameter, represents the Sigmoid function, represents the dot product, tanh is the hyperbolic tangent function;

[0029] The formula for the second intermediate layer is as follows:

[0030] ;

[0031] in Indicates the first layer The second facet feature of the facets, Representation and Facets A collection of facets with high associations. represents the first plane recognition weight matrix of layer l, is the face element u of the lth layer The attention weight is calculated by the following formula:

[0032] ;

[0033] in is a learnable attention vector, represents the vector concatenation operation, represents the second plane recognition weight matrix of layer l, represents transpose;

[0034] and Respectively represent the first The second facet feature of the facets, , F represents the total number of layers, when l=1 , , hour is equal to the second identification feature of the face element v; and Respectively represent The second feature representation of the object indexed by the uth and uth facets;

[0035] The formula for the third intermediate layer is as follows:

[0036] ;

[0037] in Represents the plane data The third identification feature of the t-th data unit of the first serial data stream corresponding to the bin, Indicates The t-1th third recognition feature of the facet, represents the tth data unit of the serial data stream No. 1 of the vth bin, and are the third and fourth serial recognition weight matrices, is the second serial recognition bias parameter, ReLU is the ReLU activation function;

[0038] The formula for the fourth intermediate layer is as follows:

[0039] ;

[0040] Indicates the first layer The fourth facet feature of the facets is represents the fourth panel feature of the u-th panel at the l-1th layer, and They represent the set of facets with high correlation with facet v and facet u respectively, represents the cardinality of a set, represents the fourth plane recognition weight matrix of layer l, , E represents the total number of layers, when l=1 , The fourth feature representation of the object representing the uth facet index, when l=E is equal to the fourth identification feature of the face element v, is the sigmoid function;

[0041] The formula for the fifth intermediate layer is as follows:

[0042] ;

[0043] in represents the tth fifth identification feature, represents the t-1th fifth identification feature, represents the tth data unit of the second serial data stream, and are the fifth and sixth serial recognition weight matrices, is the third serial identification bias parameter;

[0044] The formula for the second feature fusion layer is as follows:

[0045] ;

[0046] in, Indicates the sixth fusion state, Indicates The fourth identification feature of the plane data of the vth surface element of the layer, The set of cells representing all plane data, Indicates The fifth identifying feature, represents the total number of data items of the second serial data stream, Represents feature fusion function (splicing function);

[0047] The formula for the first feature fusion layer is as follows:

[0048] ;

[0049] in, Indicates the seventh fusion state, Indicates The second identification feature of the plane data of the vth surface element of the layer.

[0050] Furthermore, the formula of the first FC layer is as follows:

[0051] ;

[0052] in represents the output vector, and its cth component value represents the probability value of the cth strategy. The strategy with the largest probability value is selected as the output. The strategy group contains all executable strategies. A strategy includes the parameters of the spindle system, feed system, tool magazine and tool changing system, cooling and lubrication system, workpiece clamping and positioning system, and on-machine measurement system. FC represents full connectivity;

[0053] The formula for the second FC layer is as follows:

[0054] ;

[0055] in Represents the risk vector, the dth component of the risk vector represents the probability that the intermediate processing plan belongs to the dth risk level, and the risk level with the highest probability is output.

[0056] Furthermore, the training steps of the decision model are as follows:

[0057] Step 301, copy the decision model to generate an Actor network and a TargetActor network, use a feedforward neural network as a Critic network, copy the Critic network to obtain a TargetCritic network, and randomly initialize the parameters of the Actor network, TargetActor network, Critic network, and TargetCritic network;

[0058] Step 302, input The processing data of CNC machine tools collected Go to the Actor network to get the The strategy of execution ;

[0059] No. The strategy of execution Add random noise to get The noise adding strategy ;

[0060] The first The strategy of execution Input into the environment and get immediate rewards from feedback and The processing data of CNC machine tools collected ; Experience Deposit into experience pool;

[0061] Step 303, randomly obtain N experiences from the experience pool;

[0062] enter Go to the TargetActor network to get the The strategy of execution ;

[0063] Step 304, calculate the first loss value and the second loss value , gradient descent method updates the Actor network and Critic network;

[0064] ;

[0065] ;

[0066] ;

[0067] in Represents input and The output of the Critic network is Indicates that and Input the output of the Critic network, Indicates execution The comprehensive Q value when express and The output of the TargetCritic network;

[0068] ;

[0069] in, , are the first and second weight parameters, It represents the material removal rate per unit time, The calculation formula is:

[0070] ;

[0071] in, Indicates the spindle speed, Indicates the feed speed, Indicates cutting depth;

[0072] Indicates the tool wear per unit time, The calculation formula is:

[0073] ;

[0074] in, Indicates the wear of the tool. Indicates the cutting time, The calculation function of the wear rate is related to the spindle speed. , Feed speed , cutting depth , Material properties , Processing Technology and tool geometry related, The calculation formula is:

[0075] ;

[0076] in, , , , , , and are the first, second, third, fourth, fifth, sixth and seventh fitting coefficients, is a natural constant;

[0077] is the discount factor, which is a value between 0 and 1, used to balance the weights of immediate rewards and future rewards. The default value is 0.6;

[0078] pass Update the Actor network by Update the Critic network;

[0079] Step 305, every fixed number of network updates, update the TargetActor network so that its parameters are the same as those of the current Actor network, and update the TargetCritic network so that its parameters are the same as those of the current Critic network;

[0080] Step 306, until the Actor network and the Critic network converge, the step is terminated, or after the number of loops of steps 303-305 reaches a set value, the Actor network replaces the decision model and the step is terminated.

[0081] Further, the steps of transfer learning are as follows:

[0082] Step 401, load the decision model and its parameters ;

[0083] Step 402, using the collected processing task data of the actual CNC machine tool processing process as input and retraining the decision model;

[0084] Step 403: Update model parameters using gradient descent algorithm , and get the fine-tuned decision model parameters ;

[0085] ;

[0086] in, is the regularized phase, is a regularization hyperparameter, the default value is 0.01, , represents the square of the 2-norm, Indicates Minimized The value of , represents the first loss value of the new decision model, Represents the second loss value of the new decision model;

[0087] Step 404: Output the new migration model after fine-tuning , as a decision model that is ultimately applied to the actual CNC machine tool machining process.

[0088] The present invention provides a system for predicting the health status of equipment based on the large model AGENT capability, which is used to store computer-readable instructions. When the computer-readable instructions are read, the aforementioned method for predicting the health status of equipment based on the large model AGENT capability can be executed.

[0089] The beneficial effect of the present invention is that: the present invention transfers the existing processing experience and knowledge of CNC machine tools to new processing tasks by regularly collecting processing data of CNC machine tools. Compared with achieving the desired processing effect through a large number of experiments and parameter adjustments, the system can improve processing efficiency while ensuring safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0090] Figure 1 It is a flow chart of a method for predicting the health status of equipment based on the large model AGENT capability of the present invention. DETAILED DESCRIPTION

[0091] The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that the discussion of these embodiments is only to enable those skilled in the art to better understand and implement the subject matter described herein, and the functions and arrangements of the elements discussed may be changed without departing from the scope of protection of the contents of this specification. Each example may omit, replace or add various processes or components as needed. In addition, the features described relative to some examples may also be combined in other examples.

[0092] At least one embodiment of the present invention discloses a method for predicting the health status of equipment based on the ability of a large model AGENT, such as Figure 1 As shown, the following steps are included:

[0093] Step 100, acquiring the processing data of the CNC machine tool. The processing data of the CNC machine tool is collected once at each time sampling point, including the processing data of the spindle system, the feeding system, the tool magazine and tool changing system, the cooling and lubrication system, the workpiece clamping and positioning system, the on-machine measurement system and the processing task data of the CNC machine tool;

[0094] The spindle system processing data includes spindle speed, spindle torque, spindle power, spindle temperature and spindle vibration data;

[0095] The feed system processing data includes feed speed, cutting depth, cutting width, X, Y, Z axis coordinates and A, B, C axis angle data;

[0096] The processing data of tool magazine and tool changing system include tool type, tool material, tool geometry parameters, tool wear status and tool changing time data;

[0097] The processing data of cooling and lubrication system include coolant temperature, coolant pressure, coolant flow, lubricating oil temperature and lubricating oil pressure data;

[0098] The workpiece clamping and positioning system processing data includes chuck pressure, jaw position, workpiece positioning accuracy and workpiece clamping force data;

[0099] The processing data of the on-machine measurement system includes the probe trigger signal, probe position coordinates, probe pressure and measurement result data;

[0100] The processing task data of CNC machine tools include material properties, processing technology, processing procedures, workpiece drawings, processing time, and processing quality requirements;

[0101] The time sampling point refers to the moment when data is collected at a predetermined time interval;

[0102] In one embodiment of the present invention, the default value of the predetermined time interval is 1 minute.

[0103] Step 200, constructing plane data based on the processing data of the numerical control machine tool, including facets for storing data, a first serial data stream of a facet indexing a data object, and the plane data further including a relationship list for storing association relationships between data objects indexed by the facets, wherein the element value of the i-th row and j-th column of the relationship list represents a quantitative value of the association relationship between the data objects indexed by the i-th and j-th facets;

[0104] The association relationships include high and low, corresponding to two quantitative values;

[0105] Data objects include spindle system, feed system, tool magazine and tool changing system, cooling and lubrication system, workpiece clamping and positioning system, on-machine measurement system, and processing tasks;

[0106] The data objects with high correlation are defined as follows: there is a relationship between the spindle system and the feed system, cooling and lubrication system, tool magazine and tool change system; there is a relationship between the feed system and the workpiece clamping and positioning system, and the on-machine measurement system; there is a relationship between the processing task and all other data objects;

[0107] A serial data stream No. 1 of a data object includes n data units sorted in time, and the t-th data unit represents the processing data of the CNC machine tool collected at the t-th moment of the data object.

[0108] Step 300, inputting the plane data into a decision model, the decision model comprising a backbone network, the backbone network comprising a first intermediate layer, a second intermediate layer, a third intermediate layer, a fourth intermediate layer, a fifth intermediate layer, a first FC (fully connected layer) layer and a second FC layer;

[0109] The first serial data stream is input into the first intermediate layer, and the first identification feature is output to the second intermediate layer. The second intermediate layer also inputs the relationship list, and the second intermediate layer outputs the second identification feature; the first serial data stream is input into the third intermediate layer, and the third identification feature is output to the fourth intermediate layer. The fourth intermediate layer also inputs the relationship list, and the fourth intermediate layer outputs the fourth identification feature to the second feature fusion layer. The second feature fusion layer outputs the sixth fusion state to the first feature fusion layer; the first feature fusion layer inputs the second identification feature and the sixth fusion state, and outputs the seventh fusion state to the first FC layer; the second FC layer inputs the sixth fusion state and outputs the risk level;

[0110] In the first running step, the second feature fusion layer inputs the fourth recognition feature and a default state in which both components are zero values, and the size of the default state is the same as the fifth hidden state. The first FC layer outputs the initial processing plan, which includes the processing parameters of the spindle system, the feed system, the tool magazine and the tool changing system, the cooling and lubrication system, the workpiece clamping and positioning system, and the on-machine measurement system, including the spindle speed setting value, the spindle torque limit value, the spindle power limit value, the spindle temperature alarm value, the feed speed setting value, the cutting depth setting value, the cutting width setting value, the motion instructions of the X, Y, Z axis coordinates and the A, B, C axis angles, the tool selection instructions, the tool changing instructions, the tool compensation parameters, the coolant temperature setting value, the coolant pressure setting value, the coolant flow setting value, the lubricating oil temperature setting value, the lubricating oil pressure setting value, the chuck pressure setting value, the jaw position setting value, the workpiece positioning coordinate setting value, the workpiece clamping force setting value, the probe selection instruction, the measurement instruction, and the measurement coordinate setting value;

[0111] In the second operation step, the processing plan outputted by the first FC layer for the last time is used to generate a second serial data stream, and a data unit of the second serial data stream represents a processing parameter at a moment in the processing plan outputted by the first FC layer for the last time; the second serial data stream is inputted into the fifth intermediate layer, and the fifth recognition feature is outputted to the second feature fusion layer; the first FC layer outputs the intermediate processing plan, and if the risk level outputted by the second FC layer is low risk, the initial processing plan is outputted as the final processing plan; if the risk level is medium risk, the intermediate processing plan is outputted as the final processing plan; if the risk level is high risk, the second operation step is run again.

[0112] In one embodiment of the present invention, the formula of the first intermediate layer is as follows:

[0113]

[0114]

[0115]

[0116] in Represents the plane data The first identification feature of the t-th data unit of the serial data stream No. 1 corresponding to the facet, Indicates The t-1th first recognition feature of the face element, and Respectively represent The first and second intermediate features of the facet, Indicates The tth data unit of the serial data stream No. 1 of bins, and are the first and second serial recognition weight matrices, is the first serial identification bias parameter, represents the Sigmoid function, represents the dot product, and tanh is the hyperbolic tangent function.

[0117] In one embodiment of the present invention, the formula of the second intermediate layer is as follows:

[0118]

[0119] in Indicates the first layer The second facet feature of the facets, Representation and Facets A collection of facets with high associations. represents the first plane recognition weight matrix of layer l, is the face element u of the lth layer The attention weight is calculated by the following formula:

[0120]

[0121] in is a learnable attention vector, represents the vector concatenation operation, represents the second plane recognition weight matrix of layer l, represents transpose;

[0122] and Respectively represent the first The second facet feature of the facets, , F represents the total number of layers, when l=1 , , hour is equal to the second identification feature of the face element v; and Respectively represent The second feature representation of the object indexed by the uth and uth facets.

[0123] In one embodiment of the present invention, the formula of the third intermediate layer is as follows:

[0124]

[0125] in Represents the plane data The third identification feature of the t-th data unit of the first serial data stream corresponding to the bin, Indicates The t-1th third recognition feature of the facet, represents the tth data unit of the serial data stream No. 1 of the vth bin, and are the third and fourth serial recognition weight matrices, is the second serial identification bias parameter, and ReLU is the ReLU activation function.

[0126] In one embodiment of the present invention, the formula of the fourth intermediate layer is as follows:

[0127]

[0128] Indicates the first layer The fourth facet feature of the facets is represents the fourth panel feature of the u-th panel at the l-1th layer, and They represent the set of facets with high correlation with facet v and facet u respectively, represents the cardinality of a set, represents the fourth plane recognition weight matrix of layer l, , E represents the total number of layers, when l=1 , The feature representation of the object indexed by the uth facet, when l=E is equal to the fourth identification feature of the face element v, is the sigmoid function;

[0129] In one embodiment of the present invention, the formula of the fifth intermediate layer is as follows:

[0130]

[0131] in represents the tth fifth identification feature, represents the t-1th fifth identification feature, represents the tth data unit of the second serial data stream, and are the fifth and sixth serial recognition weight matrices, is the third serial identification bias parameter.

[0132] In one embodiment of the present invention, the formula of the second feature fusion layer is as follows:

[0133]

[0134] in, Indicates the sixth fusion state, Indicates The fourth identification feature of the plane data of the vth surface element of the layer, The set of cells representing all plane data, Indicates The fifth identifying feature, represents the total number of data items of the second serial data stream, Represents the feature fusion function (splicing function).

[0135] In one embodiment of the present invention, the formula of the first feature fusion layer is as follows:

[0136]

[0137] in, Indicates the seventh fusion state, Indicates The second identification feature of the plane data of the vth surface element of the layer.

[0138] In one embodiment of the present invention, the formula of the first FC layer is as follows:

[0139]

[0140] in Represents the output vector, and its cth component value represents the probability value of the cth strategy. The strategy with the largest probability value is selected as the output. The strategy group contains all executable strategies. A strategy includes the parameters of the spindle system, feed system, tool magazine and tool changing system, cooling and lubrication system, workpiece clamping and positioning system, and on-machine measurement system. FC represents full connectivity.

[0141] In one embodiment of the present invention, the formula of the second FC layer is as follows:

[0142]

[0143] in Represents the risk vector, the dth component of the risk vector represents the probability that the intermediate processing plan belongs to the dth risk level, and the risk level with the highest probability is output.

[0144] In one embodiment of the present invention, the second FC layer outputs the loss function of the risk vector It is expressed as:

[0145]

[0146] in, represents the loss value, represents the total number of classification categories, represents the total number of training samples, It is The symbol value of the training sample, if the The true classification of the training samples is The value of a classification category is 1, otherwise it is 0. is the first The training samples belong to The probability of the classification category, Represents the logarithmic function with the natural constant e as the base.

[0147] In one embodiment of the present invention, the training steps of the decision model are as follows:

[0148] Step 301, copy the decision model to generate an Actor network and a TargetActor network, use a feedforward neural network as a Critic network, copy the Critic network to obtain a TargetCritic network, and randomly initialize the parameters of the Actor network, TargetActor network, Critic network, and TargetCritic network;

[0149] Step 302, input The processing data of CNC machine tools collected Go to the Actor network to get the The strategy of execution ;

[0150] No. The strategy of execution Add random noise to get The noise adding strategy ;

[0151] The first The strategy of execution Input into the environment and get immediate rewards from feedback and The processing data of CNC machine tools collected ; Experience Deposit into experience pool;

[0152] Step 303, randomly obtain N experiences from the experience pool;

[0153] enter Go to the TargetActor network to get the The strategy of execution ;

[0154] Step 304, calculate the first loss value and the second loss value , gradient descent method updates the Actor network and Critic network;

[0155]

[0156]

[0157]

[0158] in Represents input and The output of the Critic network is Indicates that and Input the output of the Critic network, Indicates execution The comprehensive Q value when express and The output of the TargetCritic network;

[0159]

[0160] in, , are the first and second weight parameters, It represents the material removal rate per unit time, The calculation formula is:

[0161]

[0162] in, Indicates the spindle speed, Indicates the feed speed, Indicates cutting depth;

[0163] Indicates the tool wear per unit time, The calculation formula is:

[0164]

[0165] in, Indicates the wear of the tool. Indicates the cutting time, The calculation function of the wear rate is related to the spindle speed. , Feed speed , cutting depth , Material properties , Processing Technology and tool geometry related, The calculation formula is:

[0166]

[0167] in, , , , , , and are the first, second, third, fourth, fifth, sixth and seventh fitting coefficients, is a natural constant;

[0168] is the discount factor, which is a value between 0 and 1, used to balance the weights of immediate rewards and future rewards. The default value is 0.6;

[0169] pass Update the Actor network by Update the Critic network;

[0170] Step 305, every fixed number of network updates, update the TargetActor network so that its parameters are the same as those of the current Actor network, and update the TargetCritic network so that its parameters are the same as those of the current Critic network;

[0171] Step 306, until the Actor network and the Critic network converge, the step is terminated, or after the number of loops of steps 303-305 reaches a set value, the Actor network replaces the decision model and the step is terminated.

[0172] Step 400, using transfer learning to fine-tune the parameters of the decision model, the fine-tuned decision model is used in the actual CNC machine tool processing process, the steps are as follows:

[0173] Step 401, load the decision model and its parameters ;

[0174] Step 402, using the collected processing task data of the actual CNC machine tool processing process as input and retraining the decision model;

[0175] Step 403: Update model parameters using gradient descent algorithm , and get the fine-tuned decision model parameters ;

[0176]

[0177] in, is the regularized phase, is a regularization hyperparameter, the default value is 0.01, , represents the square of the 2-norm, Indicates Minimized The value of , represents the first loss value of the new decision model, Represents the second loss value of the new decision model;

[0178] Step 404: Output the new migration model after fine-tuning , as a decision model that is ultimately applied to the actual CNC machine tool machining process.

[0179] In at least one embodiment of the present invention, a system for predicting the health status of a device based on the capability of a large model AGENT is provided, which is used to store computer-readable instructions, and when the computer-readable instructions are read, the aforementioned method for predicting the health status of a device based on the capability of a large model AGENT can be executed.

[0180] The above describes an embodiment of the present invention, but this embodiment is not limited to the above-mentioned specific implementation mode. The above-mentioned specific implementation mode is merely illustrative and not restrictive. Under the guidance of this embodiment, ordinary technicians in this field can also make more forms of equivalent embodiments, all of which are within the protection of this embodiment.

Claims

1. A method for predicting equipment health status based on large model AGENT capability, characterized in that: include: Step 100, acquiring the processing data of the CNC machine tool. The processing data of the CNC machine tool is collected once at each time sampling point, including the processing data of the spindle system, the feeding system, the tool magazine and tool changing system, the cooling and lubrication system, the workpiece clamping and positioning system, the on-machine measurement system and the processing task data of the CNC machine tool; Step 200, constructing plane data based on the processing data of the numerical control machine tool, including facets for storing data, a first serial data stream of a facet indexing a data object, and the plane data further including a relationship list for storing association relationships between data objects indexed by the facets, wherein the element value of the i-th row and j-th column of the relationship list represents a quantitative value of the association relationship between the data objects indexed by the i-th and j-th facets; The association relationships include high and low, corresponding to two quantitative values; Data objects include spindle system, feed system, tool magazine and tool changing system, cooling and lubrication system, workpiece clamping and positioning system, on-machine measurement system, and processing tasks; A serial data stream No. 1 of a data object includes n data units ordered by time, and the t-th data unit represents the processing data of the CNC machine tool collected at the t-th moment of the data object; Step 300, inputting the plane data into a decision model, the decision model comprising a backbone network, the backbone network comprising a first intermediate layer, a second intermediate layer, a third intermediate layer, a fourth intermediate layer, a fifth intermediate layer, a first FC layer and a second FC layer, the first FC layer outputting a processing plan of a CNC machine tool, and the second FC layer outputting a risk level of the processing plan; Step 400, using transfer learning to fine-tune the parameters of the decision model, and the fine-tuned decision model is used in the actual CNC machine tool processing process.

2. A method for predicting equipment health status based on large model AGENT capability according to claim 1, characterized in that: The time sampling point refers to the moment when data is collected at a predetermined time interval. The default value of the predetermined time interval is 1 minute.

3. The method for predicting the health status of equipment based on the large model AGENT capability according to claim 1 is characterized in that: The spindle system processing data includes spindle speed, spindle torque, spindle power, spindle temperature and spindle vibration data; The feed system processing data includes feed speed, cutting depth, cutting width, X, Y, Z axis coordinates and A, B, C axis angle data; The processing data of tool magazine and tool changing system include tool type, tool material, tool geometry parameters, tool wear status and tool changing time data; The processing data of cooling and lubrication system include coolant temperature, coolant pressure, coolant flow, lubricating oil temperature and lubricating oil pressure data; The workpiece clamping and positioning system processing data includes chuck pressure, jaw position, workpiece positioning accuracy and workpiece clamping force data; The processing data of the on-machine measurement system includes the probe trigger signal, probe position coordinates, probe pressure and measurement result data; The processing task data of CNC machine tools include material properties, processing technology, processing procedures, workpiece drawings, processing time, and processing quality requirements.

4. The method for predicting the health status of equipment based on the large model AGENT capability according to claim 1 is characterized in that: Data objects with high correlation are defined as follows: there is a relationship between the spindle system and the feed system, cooling and lubrication system, tool magazine and tool changing system; there is a relationship between the feed system and the workpiece clamping and positioning system, and the on-machine measurement system; there is a relationship between the processing task and all other data objects.

5. The method for predicting the health status of equipment based on the large model AGENT capability according to claim 1 is characterized in that: Inputting serial data stream No. 1 into the first intermediate layer, outputting the first identification feature to the second intermediate layer, the second intermediate layer also inputting the relationship list, and the second intermediate layer outputting the second identification feature; The first serial data stream is input into the third intermediate layer, and the third identification feature is output to the fourth intermediate layer. The fourth intermediate layer also inputs the relationship list. The fourth intermediate layer outputs the fourth identification feature to the second feature fusion layer. The second feature fusion layer outputs the sixth fusion state to the first feature fusion layer. The first feature fusion layer inputs the second recognition feature and the sixth fusion state, and outputs the seventh fusion state to the first FC layer; the second FC layer inputs the sixth fusion state and outputs the risk level; In the first running step, the second feature fusion layer inputs the fourth recognition feature and a default state in which both components are zero values, and the size of the default state is the same as the fifth hidden state. The first FC layer outputs the initial processing plan, which includes the processing parameters of the spindle system, the feed system, the tool magazine and the tool changing system, the cooling and lubrication system, the workpiece clamping and positioning system, and the on-machine measurement system, including the spindle speed setting value, the spindle torque limit value, the spindle power limit value, the spindle temperature alarm value, the feed speed setting value, the cutting depth setting value, the cutting width setting value, the motion instructions of the X, Y, Z axis coordinates and the A, B, C axis angles, the tool selection instructions, the tool changing instructions, the tool compensation parameters, the coolant temperature setting value, the coolant pressure setting value, the coolant flow setting value, the lubricating oil temperature setting value, the lubricating oil pressure setting value, the chuck pressure setting value, the jaw position setting value, the workpiece positioning coordinate setting value, the workpiece clamping force setting value, the probe selection instruction, the measurement instruction, and the measurement coordinate setting value; In the second operation step, the processing plan outputted by the first FC layer for the last time is used to generate a second serial data stream, and a data unit of the second serial data stream represents a processing parameter at a moment in the processing plan outputted by the first FC layer for the last time; the second serial data stream is inputted into the fifth intermediate layer, and the fifth recognition feature is outputted to the second feature fusion layer; the first FC layer outputs the intermediate processing plan, and if the risk level outputted by the second FC layer is low risk, the initial processing plan is outputted as the final processing plan; if the risk level is medium risk, the intermediate processing plan is outputted as the final processing plan; if the risk level is high risk, the second operation step is run again.

6. A method for predicting equipment health status based on large model AGENT capability according to claim 5, characterized in that: The formula for the first intermediate layer is as follows: ； ； ； in Represents the plane data The first identification feature of the t-th data unit of the serial data stream No. 1 corresponding to the facet, Indicates The t-1th first recognition feature of the face element, and Respectively represent The first and second intermediate features of the facet, Indicates The tth data unit of the serial data stream No. 1 of bins, and are the first and second serial recognition weight matrices, is the first serial identification bias parameter, represents the Sigmoid function, represents the dot product, tanh is the hyperbolic tangent function; The formula for the second intermediate layer is as follows: ； in Represents the first layer The second facet feature of the facet, Representation and Facets A collection of facets with high associations. represents the first plane recognition weight matrix of layer l, is the face element u of the lth layer The attention weight is calculated by the following formula: ； in is a learnable attention vector, represents the vector concatenation operation, represents the second plane recognition weight matrix of layer l, represents transpose; and Respectively represent the first The second facet feature of the facet, , F represents the total number of layers, when l=1 , , hour is equal to the second identification feature of the face element v; and Respectively represent The second feature representation of the object indexed by the uth and uth facets; The formula for the third intermediate layer is as follows: ； in Represents the plane data The third identification feature of the t-th data unit of the first serial data stream corresponding to the bin, Indicates The t-1th third recognition feature of the facet, represents the tth data unit of the serial data stream No. 1 of the vth bin, and are the third and fourth serial recognition weight matrices, is the second serial recognition bias parameter, ReLU is the ReLU activation function; The formula for the fourth intermediate layer is as follows: ； Represents the first layer The fourth facet feature of the facets is represents the fourth panel feature of the u-th panel at the l-1th layer, and They represent the set of facets with high correlation with facet v and facet u respectively, represents the cardinality of a set, represents the fourth plane recognition weight matrix of layer l, , E represents the total number of layers, when l=1 , The fourth feature representation of the object representing the uth facet index, when l=E is equal to the fourth identification feature of the face element v, is the sigmoid function; The formula for the fifth intermediate layer is as follows: ； in represents the tth fifth identification feature, represents the t-1th fifth identification feature, represents the tth data unit of the second serial data stream, and are the fifth and sixth serial recognition weight matrices, is the third serial identification bias parameter; The formula for the second feature fusion layer is as follows: ； in, Indicates the sixth fusion state, Indicates The fourth identification feature of the plane data of the vth surface element of the layer, The set of cells representing all plane data, Indicates The fifth identifying feature, represents the total number of data items of the second serial data stream, represents the feature fusion function; The formula for the first feature fusion layer is as follows: ； in, Indicates the seventh fusion state, Indicates The second identification feature of the plane data of the vth bin of the layer.

7. A method for predicting equipment health status based on large model AGENT capability according to claim 6, characterized in that: The formula for the first FC layer is as follows: ； in represents the output vector, and its cth component value represents the probability value of the cth strategy. The strategy with the largest probability value is selected as the output. The strategy group contains all executable strategies. A strategy includes the parameters of the spindle system, feed system, tool magazine and tool changing system, cooling and lubrication system, workpiece clamping and positioning system, and on-machine measurement system. FC represents full connectivity; The formula for the second FC layer is as follows: ； in Represents the risk vector, the dth component of the risk vector represents the probability that the intermediate processing plan belongs to the dth risk level, and the risk level with the highest probability is output.

8. The method for predicting the health status of equipment based on the large model AGENT capability according to claim 1 is characterized in that: The training steps of the decision model are as follows: Step 301, copy the decision model to generate an Actor network and a TargetActor network, use a feedforward neural network as a Critic network, copy the Critic network to obtain a TargetCritic network, and randomly initialize the parameters of the Actor network, TargetActor network, Critic network, and TargetCritic network; Step 302, input The processing data of CNC machine tools collected Go to the Actor network to get the The strategy of execution ; No. The strategy of execution Add random noise to get The noise adding strategy ; The first The strategy of execution Input into the environment and get immediate rewards from feedback and The processing data of CNC machine tools collected ; Experience Deposit into experience pool; Step 303, randomly obtain N experiences from the experience pool; enter Go to the TargetActor network to get the The strategy of execution ; Step 304, calculate the first loss value and the second loss value , gradient descent method updates the Actor network and Critic network; ； ； ； in Represents input and The output of the Critic network is Indicates that and Input the output of the Critic network, Indicates execution The comprehensive Q value when express and The output of the TargetCritic network; ； in, , are the first and second weight parameters, It represents the material removal rate per unit time, The calculation formula is: ； in, Indicates the spindle speed, Indicates the feed speed, Indicates cutting depth; Indicates the tool wear per unit time, The calculation formula is: ； in, Indicates the wear of the tool. Indicates the cutting time, The calculation function of the wear rate is related to the spindle speed. , Feed speed , cutting depth , Material properties , Processing Technology and tool geometry related, The calculation formula is: ； in, , , , , , and are the first, second, third, fourth, fifth, sixth and seventh fitting coefficients, is a natural constant; is the discount factor, which is a value between 0 and 1, used to balance the weights of immediate rewards and future rewards. The default value is 0.6; pass Update the Actor network by Update the Critic network; Step 305, every fixed number of network updates, update the TargetActor network so that its parameters are the same as those of the current Actor network, and update the TargetCritic network so that its parameters are the same as those of the current Critic network; Step 306, until the Actor network and the Critic network converge, the step is terminated, or after the number of loops of steps 303-305 reaches a set value, the Actor network replaces the decision model and the step is terminated.

9. The method for predicting the health status of equipment based on the large model AGENT capability according to claim 1 is characterized in that: The steps of transfer learning are as follows: Step 401, load the decision model and its parameters ; Step 402, using the collected processing task data of the actual CNC machine tool processing process as input and retraining the decision model; Step 403: Update model parameters using gradient descent algorithm , and obtain the fine-tuned decision model parameters ; ； in, is the regularized phase, is a regularization hyperparameter, the default value is 0.01, , represents the square of the 2-norm, Indicates Minimized The value of , represents the first loss value of the new decision model, Represents the second loss value of the new decision model; Step 404: Output the new migration model after fine-tuning , as a decision model that is ultimately applied to the actual CNC machine tool machining process.

10. A system for predicting equipment health status based on large model AGENT capability, characterized in that: It is used to store computer-readable instructions, and when the computer-readable instructions are read, it can execute a method for predicting the health status of equipment based on the large model AGENT capability as described in any one of claims 1-9.

Images