Composite material cutting machine control method and system

Abstract

The invention discloses a composite material cutting machine control method and system, and relates to the technical field of intelligent manufacturing and industrial process control, and the method comprises the steps: collecting cutting process data through a multi-mode sensor, and extracting features; establishing a tool wear model based on a weakening layer theory, and coupling mechanical and thermochemical wear effects; predicting the cutting quality and the residual life of the cutter; a near-end strategy optimization algorithm assisted by a physical model is adopted to generate control parameters, and the operation safety is ensured through a safety projection layer; and finally, a closed-loop control and online updating system is formed. According to the method, multi-target optimization control over the composite material cutting process is achieved, and the machining quality, the production efficiency and the tool utilization rate are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent manufacturing and industrial process control, and particularly relates to a composite material cutting machine control method and system. BACKGROUND

[0002] In the composite material cutting process, tool wear is a key factor affecting the processing quality, efficiency and cost. Traditional control methods mostly use fixed parameters or adjustment based on simple rules, which are difficult to cope with the time-varying characteristics of the processing process and the coupling effect of multiple wear modes. Existing data-based intelligent control methods often lack deep integration with physical mechanisms, and have problems such as lack of explainability of decision-making, insufficient consideration of long-term benefits, and possible violation of physical safety constraints in the exploration process.

[0003] Some existing adaptive control technologies, such as the "adaptive control of composite lay-up cutting" described in CN101990485B, although can generate feedback control signals by sensing tool parameters to optimize the feed rate, still lack the ability to model the deep physical mechanism of tool health state and multi-objective optimization. SUMMARY

[0004] In order to solve the technical problems in the prior art, the present application provides a composite material cutting machine control method and system.

[0005] The present application is realized by the following technical solutions:

[0006] A composite material cutting machine control method, comprising:

[0007] S1: multi-modal sensor data acquisition and feature extraction, the multi-modal sensor data acquisition includes acquisition of cutting force signals, cutting speed signals, vibration signals, acoustic emission signals, infrared thermal images and machine parameters;

[0008] S2: tool wear model construction based on multi-mode coupling, including estimation through interface temperature and mechanical energy input, and updating of weakened layer state based on this, so as to calculate the wear amount;

[0009] S3: intelligent prediction of cutting quality and remaining life;

[0010] S4: adaptive adjustment of control parameters based on a near-end policy optimization model assisted by a physical model;

[0011] S5: closed-loop execution and model online updating.

[0012] Further, the tool wear model construction based on multi-mode coupling further includes establishing a weakened layer growth kinetics model, which is combined with thermochemical and mechanical wear calculation.

[0013] Further, the weakening layer growth kinetics model expression is as follows:

[0014]

[0015] Wherein, S(t) is the weakening layer thickness of the tool surface at time t; is the thermo-chemical wear, wherein, is the activation energy (determined by the tool coating and the workpiece material), R is the gas constant, and A is a proportional constant, is the interface temperature; is the saturation term, and is is the maximum possible thickness; is the mechanical wear term; is the peeling coefficient.

[0016] Further, the physical model assisted proximal policy optimization model input includes the features of the data collected by the multi-modal sensor and the current machine parameters, forming a current state vector, and the model output is the control parameter at the next moment, including the spindle speed, the feed speed and the cutting depth per unit time.

[0017] Further, the physical model assisted proximal policy optimization model includes an Actor network and a Critic network.

[0018] Further, the physical model assisted proximal policy optimization model includes a safety projection layer added after the output of the Actor network in the PPO model, and the safety projection layer is a neural network layer used for control adjustment of the output control parameter, and projects the original control parameter to the nearest safety control parameter.

[0019] Further, the safety projection layer receives the original control parameter of the Actor and the current state, quickly calculates the gradient of the constraint function using a simplified physical model, constructs a constraint matrix and a vector, constructs a QP problem: the goal is to minimize the control parameter correction amplitude, and solves the built-in QP solver to calculate the optimal safety control parameter in real time.

[0020] Further, the Critic network obtains the reward and the state to calculate the state value as the basis for updating the parameters of the Actor network and the Critic network; the reward includes immediate reward, long-term punishment and physical constraint.

[0021] The application also provides a composite material cutting machine control system based on the composite material cutting machine control method as described above, which comprises:

[0022] A multi-modal sensor data acquisition and feature extraction module is used for multi-modal sensor data acquisition and feature extraction.

[0023] a tool wear model construction module for establishing a wear evaluation model fusing a physical mechanism, wherein interface temperature and mechanical energy input estimation are included, and the weakened layer state is updated based on this to calculate the wear amount;

[0024] a cutting quality and residual life intelligent prediction module for calculating the prediction of cutting quality and residual life;

[0025] a control parameter determination module based on a physical model assisted proximal policy optimization model combined with a safety projection layer to determine the control parameters.

[0026] In addition, in order to achieve the above-mentioned purpose, the application also provides a computer readable storage medium, and the computer readable storage medium stores the program instructions of the composite material cutting machine control method, and the program instructions of the composite material cutting machine control method can be executed by one or more processors to realize the steps of the composite material cutting machine control method as described above.

[0027] Compared with the prior art, the application has the following beneficial effects:

[0028] (1) Mechanism and data driven fusion are realized, the agent has a deeper understanding of the wear mechanism by introducing the weakened layer physical model, and the explainability and accuracy of the decision are improved.

[0029] (2) Based on multi-objective optimization, the innovative reward function design balances the multi-objective requirements of quality, efficiency, tool life, etc., and the durability reward item guides the long-term planning of the control parameters.

[0030] (3) Safety guarantee is added, and the safety projection layer limits the control action in the safety range through the QP solver, avoiding dangerous behaviors in the exploration process of the traditional algorithm.

[0031] The adaptive ability is enhanced, and the online updating mechanism enables the system to adapt to the time-varying characteristics of the tool and the material, and maintain long-term control performance. BRIEF DESCRIPTION OF DRAWINGS

[0032] The drawings described herein are used to provide further understanding of the present application, and form a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application, and do not constitute an improper limitation on the present application. In the drawings:

[0033] Figure 1 is a composite material cutting machine control method flowchart according to an embodiment of the present application;

[0034] Figure 2 is a proximal policy optimization model structure based on a physical model according to an embodiment of the present application. DETAILED DESCRIPTION

[0035] The embodiments of the present application will be described in detail below with reference to the drawings.

[0036] The following detailed description of the application is provided for the purpose of understanding by those skilled in the art. It is obvious that the described embodiments are only a part of the embodiments of the present application, and are not all the embodiments. The present application can also be implemented or applied by other different specific embodiments, and the details in the specification can be modified or changed based on different views and applications without departing from the spirit of the present application. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0037] It should also be noted that the drawings provided in the following embodiments only schematically illustrate the basic concept of the present application, and only show the components related to the present application in the drawings, not the number, shape and size of the components when actually implemented. The shape, number and ratio of each component when actually implemented can be arbitrarily changed, and the layout form of the components can also be more complex.

[0038] Referring to Figure 1 A composite material cutting machine control method, comprising the following steps:

[0039] S1: Multi-modal sensor data acquisition and feature extraction

[0040] This step is the basis for the system to perceive the physical world. By deploying multiple sensors, information about the cutting process and tool state is captured from different dimensions.

[0041] S11: Multi-modal sensor data acquisition;

[0042] Multi-modal sensor data acquisition includes acquisition of cutting force signals, cutting speed signals, vibration signals, acoustic emission signals, infrared thermal images, and machine parameters.

[0043] Optionally, the cutting force signals are three-directional cutting forces (F x , F y , F z ) measured by a dynamometer, and the three-directional cutting forces are integrated into a main cutting force The force signal directly reflects the intensity of the interaction between the tool and the material, and the cutting line speed .

[0044] The vibration signal is a three-directional vibration signal (a x , ay , a z )。Vibration spectrum contains information of tool wear, chatter, etc.

[0045] The acoustic emission signal uses an acoustic emission sensor to collect a high-frequency stress wave signal (AE). Acoustic emission is extremely sensitive to micro-fracture and crack propagation of materials.

[0046] The infrared thermal image is the temperature field distribution (T(x, y, t)) of the tool-workpiece contact area monitored by an infrared thermal imager. This is the key to monitoring thermal chemical wear.

[0047] The machine parameters: spindle speed n, feed speed f, cutting depth per unit time a p .

[0048] S12: Feature extraction;

[0049] From the original sensor signal, the features strongly related to tool wear are extracted to form a feature vector X t , which includes time domain features, frequency domain features and time-frequency domain features of cutting force signals, vibration signals and high-frequency stress wave signals; temperature features of infrared thermal images.

[0050] The time domain features include mean, root mean square, variance, peak value.

[0051] The frequency domain features are obtained by Fourier transform to get the dominant frequency and its amplitude.

[0052] The time-frequency domain features are extracted by wavelet packet transform, etc. to get the energy features in a specific frequency band, which is especially effective for non-stationary signals.

[0053] The temperature features include the highest temperature T max (t), the average temperature T avg (t), and the high-temperature area A hot (t).

[0054] Finally, a high-dimensional feature vector X t is obtained by combining the above features.

[0055] S2: Tool wear model construction based on multi-mode coupling;

[0056] Most of the existing technology adopts black box model for equipment health state evaluation, which cannot effectively feedback the physical properties and other information of the equipment. Therefore, the present application establishes a wear evaluation model integrating physical mechanism, which includes interface temperature and mechanical energy input estimation, and updates the weakened layer state based on this to calculate the wear amount, specifically including the following steps:

[0057] S21: Interface temperature and mechanical energy input estimation;

[0058] Interface temperature estimation:

[0059]

[0060] where, T(t) is the interface temperature at time t, T0 is the ambient temperature, and A and E are constants and exponent calibrated by experiments, F is the main cutting force, V is the cutting linear speed.

[0061] Mechanical energy input estimation:

[0062]

[0063] M(t) is the mechanical energy input at time t.

[0064] S22: Establish the weakened layer growth kinetics model, the expression is as follows:

[0065]

[0066] where, S(t) is the thickness of the weakened layer on the tool surface at time t. This is a key endogenous variable, connecting the two wear modes; T(t) is the thermal-chemical wear, which follows the Arrhenius-type rate equation, describing the growth of the weakened layer caused by thermal-chemical reactions, where, E is the activation energy (determined by the tool coating and workpiece material), R is the gas constant, and A is a proportional constant, T(t) is the interface temperature; is a saturation term, indicating that the weakened layer cannot thicken indefinitely, and its maximum possible thickness is ; M(t) is the mechanical wear term, indicating that the mechanical wear rate will strip and remove the weakened layer. is the stripping coefficient. This means that mechanical wear will "clean up" the weakened layer, temporarily slowing down the thermal-chemical wear, but exposing the new surface will accelerate new thermal-chemical reactions.

[0067] The mechanical wear rate is not a constant, it is significantly dependent on the thickness of the weakened layer, the thicker the weakened layer, the worse the material's resistance to mechanical wear.

[0068]

[0069] is the baseline mechanical wear rate proportional to the cutting energy; is the baseline mechanical wear coefficient; (1 + η * S(t)) is the coupling term, where η is the wear sensitivity coefficient, and the weakened layer S(t) amplifies the effect of mechanical wear, the larger S(t) is, the higher the efficiency and the faster the rate of mechanical wear.

[0070] S23: Weakened layer state update

[0071] According to the weakened layer growth kinetics model, the weakened layer state update is obtained:

[0072]

[0073] S24: Wear amount calculation and decomposition

[0074] According to the updated weakened layer state , the current total wear amount is calculated and decomposed into mechanical wear and thermo-chemical wear.

[0075] Thermo-chemical wear increment calculation:

[0076]

[0077] is the thermo-chemical wear rate coefficient.

[0078] Mechanical wear increment calculation:

[0079]

[0080] Total wear amount calculation:

[0081]

[0082] The present application quantifies the weakened layer formed on the surface of the tool due to the thermal-mechanical coupling effect, revealing the internal mechanism of wear. It serves as a bridge to dynamically couple mechanical wear and thermo-chemical wear: the thermo-chemical process generates a weakened layer, and the mechanical process peels it off and accelerates tool failure. The model changes the health assessment from "black box" prediction to "white box" simulation, not only enabling more accurate prediction of tool life, but also diagnosing the current dominant wear mode. As a result, the control system can implement more predictive parameter adjustment, thereby significantly improving tool utilization and overall production efficiency while ensuring machining quality.

[0083] S3: Intelligent prediction of cutting quality and remaining life

[0084] S31: Cutting quality calculation

[0085] The cutting quality Q includes roughness, layering factor, and burr height, which is the current tool state and cutting parameters The function r is a trained machine learning model that gives both the predicted value and its uncertainty.

[0086]

[0087] The function r is a trained machine learning model that gives both the predicted value and its uncertainty.

[0088] S32: Residual life RUL prediction:

[0089] Define the failure threshold as VB max When the wear reaches this value, the quality is unqualified.

[0090] Forward inference of the wear model in step S2:

[0091]

[0092] The estimate of Delta VB combines data-driven prediction and physical model prediction. Through recursive calculation, find the minimum integer k that makes VB(t + k*Delta t) >= VB_max. This k is the predicted residual cutting number, i.e. the residual life RUL.

[0093] S4: Adaptive adjustment of control parameters based on a proximal policy optimization (PPO) model assisted by a physical model;

[0094] The model input includes features X of data collected by multi-modal sensors t and current machine parameters, forming the current state vector The model output is the control parameter at the next time, including spindle speed n, feed speed f, and cutting depth per unit time a p The model structure is shown in Figure 2 The present application adds a safety projection layer in the PPO model, which is a neural network layer located after the output of the Actor network, used to control and adjust the original control parameters, and project the original control parameters to the nearest safe control parameters.

[0095] The input of the Actor network is the state vector The output of the Actor network is the original control parameter The safety projection layer receives the original control parameter and the current state of the Actor, uses a simplified physical model to quickly calculate the gradient of the constraint function, constructs the constraint matrix A and vector b; constructs the QP problem: the objective is to minimize the correction amplitude - ||², the constraint condition is A * ≤b; solve: the built-in QP solver calculates the optimal safe control parameter in real time.

[0096] The executor executes the safety control parameters , obtains a new state , and calculates a reward ; the Critic network obtains the reward and the state , calculates the state value , and uses the output state value to update the parameters of the Actor network and the Critic network; finally, the parameters of each step , , , , are put into the experience replay buffer.

[0097] wherein the reward balances the immediate reward, the long-term penalty, and the physical constraint, wherein the immediate reward includes a quality reward and an efficiency reward, the long-term penalty includes a durability reward, and the expression is as follows:

[0098]

[0099] is the quality reward, is the efficiency reward, is the durability reward, is the physical constraint.

[0100]

[0101] wherein is the quality reward weight, is the cutting quality, is the target value of the cutting quality.

[0102] The closer the cutting quality is to the target value, the higher the reward is.

[0103]

[0104] is the efficiency reward weight, is the material removal rate, which is approximately represented by controlling the parameter ;

[0105]

[0106] wherein is the change in the predicted service life, and are the remaining service life prediction values at times t and t-1, respectively.

[0107] This term is the key point, the reward function of the present application rewards the increase of the remaining life, if the current control parameter makes the predicted total tool life extend, it will get a positive reward. This encourages the agent to adopt a strategy that not only considers the current cutting, but also benefits the long-term use, avoiding the short-sighted behavior caused by the greedy algorithm.

[0108]

[0109] wherein, 、 、 are the set main cutting force, interface temperature and cutting quality constraint values respectively.

[0110] This term is a penalty term. Once any one of the cutting force, temperature or cutting quality exceeds its safety threshold, a negative reward will be applied. is the weight coefficient of the physical constraint, which is much larger than other weights, ensuring that the agent will never learn a strategy that violates these core safety constraints.

[0111] The gradient of the constraint function is quickly calculated using a simplified physical model, and the constraint matrix A and vector b are constructed as follows:

[0112] a) Let the original nonlinear constraint expression be as follows:

[0113]

[0114] is the physical parameter obtained under state S and control parameter a, such as cutting force, temperature, etc. is the set maximum constraint value of this physical parameter;

[0115] b) At the current point , Taylor expansion is performed on to approximately calculate the physical parameter after executing the new control parameter :

[0116]

[0117] wherein, is the gradient of the physical parameter model with respect to the control parameter a, such as the cutting force model;

[0118] c) Construct a linear inequality, and bring the linearized expression into the original constraint:

[0119]

[0120] Move the term:

[0121]

[0122] The following can be obtained:

[0123]

[0124]

[0125] represents the sensitivity of the physical parameter to the change of each control parameter, represents the safety margin.

[0126] d) all the constraints , Stacked together, the complete constraint matrix A and the vector b are formed.

[0127] The advantage of the application of adding a safety projection layer is not only to ensure the safety of the action, but also to find the nearest safety control parameter a t ' is the smallest safety control parameter to be modified, which maximizes the original intention of the actor and can be applied to continuous control.

[0128] S5: closed-loop execution and model online update;

[0129] Closed-loop execution:

[0130] The optimal safety control parameter obtained by S4 optimization is issued to the PLC to control the cutting machine to perform this cutting.

[0131] Model online update:

[0132] After cutting is completed, the new round of collected sensor data X(t-1) and the actual measured tool wear VB measured , wherein VB measured can be obtained by regular visual inspection or shutdown measurement.

[0133] Use the new data to form a new training sample for model online update.

[0134] Through the five steps, the system forms a complete closed loop and realizes truly intelligent, self-adaptive and creative optimization control.

[0135] In this embodiment, mechanism and data-driven fusion is realized, and the agent has a deeper understanding of the wear mechanism by introducing a weakening layer physical model, which improves the explainability and accuracy of decision-making.

[0136] Based on multi-objective optimization, the innovative reward function design balances the multi-objective requirements of quality, efficiency, tool life, etc., and the durability reward item guides the long-term planning of the control parameter.

[0137] Security guarantee is added, and a security projection layer limits control actions in a safe range through a QP solver to avoid dangerous behaviors in the exploration process of traditional algorithms. Adaptive capability is enhanced, and an online updating mechanism enables the system to adapt to time-varying characteristics of tools and materials and maintain long-term control performance.

[0138] The embodiment of the present application also provides a composite material cutting machine control system based on the composite material cutting machine control method.

[0139] A multi-modal sensing data acquisition and feature extraction module is used for multi-modal sensing data acquisition and feature extraction.

[0140] A tool wear model construction module is used for establishing a wear evaluation model fusing a physical mechanism, which includes interface temperature and mechanical energy input estimation, and updating a weakened layer state based on the same to perform wear amount calculation.

[0141] A cutting quality and residual life intelligent prediction module is used for calculating cutting quality and residual life prediction.

[0142] A control parameter determination module determines control parameters based on a physical model assisted proximal policy optimization model combined with a security projection layer.

[0143] In addition, the embodiment of the present application also provides a computer readable storage medium, and the computer readable storage medium stores a program instruction of the composite material cutting machine control method. The program instruction of the composite material cutting machine control method can be executed by one or more processors to realize the steps of the composite material cutting machine control method.

[0144] The above-described embodiments only describe the preferred embodiments of the present application, and do not limit the scope of the present application. Without departing from the design spirit of the present application, various modifications and improvements of the technical solutions of the present application made by those skilled in the art shall fall within the protection scope of the present application defined by the claims.

Claims

1. A control method for a composite material cutting machine, characterized in that, include: S1: Multimodal sensing data acquisition and feature extraction, wherein the multimodal sensing data acquisition includes acquiring cutting force signals, cutting speed signals, vibration signals, acoustic emission signals, infrared thermography, and machine parameters; S2: Construction of tool wear model based on multi-mode coupling, including estimation through interface temperature and mechanical energy input, and updating the weakened layer state based on this, thereby calculating the wear amount; S3: Intelligent prediction of cutting quality and remaining life; S4: Adaptive adjustment of control parameters for a physics-based near-end strategy optimization model; S5: Closed-loop execution and online model updates.

2. The control method for a composite material cutting machine according to claim 1, characterized in that, The construction of the tool wear model based on multi-mode coupling also includes establishing a weakened layer growth dynamics model, which is obtained by combining thermochemical and mechanical wear calculations.

3. The control method for a composite material cutting machine according to claim 2, characterized in that, The expression for the growth kinetics model of the weakened layer is as follows: Where S(t) is the thickness of the weakened layer on the tool surface at time t; This is thermochemical wear, in which, It is the activation energy (determined by the tool coating and workpiece material), R is the gas constant, and A is a proportionality constant. Interface temperature; For saturation term, The maximum possible thickness; For mechanical wear; This is the stripping coefficient.

4. The control method for a composite material cutting machine according to claim 1, characterized in that, The input of the physical model-assisted near-end strategy optimization model includes the features of the data collected by the multimodal sensors and the machine parameters at the current moment, forming the current state vector. The model output is the control parameters for the next moment, including the spindle speed, feed rate, and cutting depth per unit time.

5. The control method for a composite material cutting machine according to claim 4, characterized in that, The physics-based near-end policy optimization model includes an Actor network and a Critic network.

6. The control method for a composite material cutting machine according to claim 5, characterized in that, The physical model-assisted proximal policy optimization model includes adding a safe projection layer after the Actor network output in the PPO model. The safe projection layer is a neural network layer used to control and adjust the output control parameters, projecting the original control parameters to the nearest safe control parameters.

7. The control method for a composite material cutting machine according to claim 6, characterized in that, The safety projection layer receives the Actor's original control parameters and current state, uses a simplified physical model to quickly calculate the gradient of the constraint function, and constructs the constraint matrix and vector; it constructs a QP problem: the goal is to minimize the control parameter correction magnitude; and it solves the optimal safety control parameters in real time using the built-in QP solver.

8. The control method for a composite material cutting machine according to claim 6, characterized in that, The Critic network obtains rewards and calculates state values ​​as the basis for updating the parameters of the Actor network and the Critic network; the rewards include immediate rewards, long-term penalties, and physical constraints.

9. A composite material cutting machine control system, based on the composite material cutting machine control method as described in any one of claims 1 to 8, comprising: A multimodal sensing data acquisition and feature extraction module, which is used for multimodal sensing data acquisition and feature extraction; The tool wear model construction module is used to establish a wear assessment model that incorporates physical mechanisms, including the estimation of interface temperature and mechanical energy input, and to update the weakened layer state based on this, thereby calculating the amount of wear. The intelligent prediction module for cutting quality and remaining life is used to calculate predictions of cutting quality and remaining life. The control parameter determination module determines the control parameters based on a physical model-assisted near-end strategy optimization model combined with a safety projection layer.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores program instructions for a composite material cutting machine control method, which can be executed by one or more processors to implement the steps of the composite material cutting machine control method as described in any one of claims 1 to 8.

Images