Alloy processing workshop scheduling method based on physical neural network real-time monitoring

Abstract

The application provides an alloy processing workshop scheduling method based on real-time monitoring of a physical neural network, and belongs to the technical field of intelligent scheduling of alloy processing workshops.A deep operator network combining physical information and a deep learning model is proposed for tool life prediction.The network can not only learn the dynamic changes of tool wear from data, but also enhance the interpretability and prediction ability of the model through physical signals.The application deeply combines tool life prediction and workshop production scheduling, and proposes an intelligent scheduling method based on tool residual life, which dynamically adjusts the allocation of processing tasks and the tool replacement plan by monitoring the tool state in real time, thereby avoiding overuse or premature replacement of tools.A hybrid optimization strategy combining an adaptive genetic algorithm and reinforcement learning is also proposed, which performs global optimization through the genetic algorithm and local optimization through reinforcement learning, dynamically responds to changes in the workshop environment, and ensures the real-time and high efficiency of the scheduling scheme.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent scheduling of alloy processing workshops, and in particular relates to an alloy processing workshop scheduling method based on real-time monitoring of a physical neural network. Background Art

[0002] The machining of alloy materials typically requires high precision and efficiency, especially when machining complex components, placing extremely high demands on the performance of CNC machine tools. CNC machine tool cutting tools are core tools in the machining process, and their service life directly impacts the quality and cost of the finished product. However, due to the high temperatures and high cutting speeds encountered by cutting tools, wear and fatigue are inevitable. This wear not only reduces machining accuracy but can also lead to tool failure or even breakage, causing irreversible damage to the equipment and product. Effectively monitoring tool condition and predicting fatigue life are key technologies for ensuring the stability and cost-effectiveness of machining processes.

[0003] In recent years, advances in sensor technology have enabled real-time acquisition of physical parameters such as machine tool vibration signals, cutting force signals, and temperature signals, providing a rich data foundation for tool wear monitoring. Simultaneously, the emergence of artificial intelligence (AI), particularly deep learning models, has provided powerful tools for extracting underlying patterns from this complex data. These technologies can leverage extensive historical data to model and predict tool wear trends, providing data support for timely tool replacement and optimized machining parameters. However, "black box" prediction methods that rely solely on deep learning models often lack physical interpretability and struggle to integrate machining mechanisms with data models, limiting their widespread adoption and application in industrial scenarios.

[0004] In addition, the coupling problem between workshop production scheduling and tool life prediction has not been effectively solved. Traditional scheduling methods are mostly based on empirical rules and lack the ability to dynamically respond to the real-time status of tools. This can easily lead to excessive use of tools or premature replacement, thereby reducing production efficiency and resource utilization. Therefore, the in-depth integration of tool life prediction and intelligent scheduling, and the optimization of processing task allocation and tool replacement plans through real-time monitoring of tool status, have become an important direction for solving the bottleneck problem of alloy processing workshops. The existing technology has the following defects:

[0005] 1. Disadvantages of the empirical formula method:

[0006] First, accuracy is low. This method relies on manually defined empirical formulas and fails to fully account for dynamic changes during the machining process. Second, flexibility is limited. It struggles to adapt to the machining conditions of different machine tools and cutting tools, especially in complex machining scenarios. Finally, resources are wasted. This can lead to premature tool replacement or overuse, reducing production efficiency.

[0007] 2. Disadvantages of data-driven approach:

[0008] Data dependency during model training. Data-driven models require a large amount of high-quality historical data as training samples. Insufficient data or low-quality data will affect model performance.

[0009] Poor model interpretability. Traditional deep learning models lack the ability to directly interpret the physical properties of cutting tools. Relying solely on data features, they may not be able to capture key information in physical signals well.

[0010] High real-time requirements: Existing deep learning models are mostly trained offline, and there is still a certain lag in the requirements for real-time monitoring and scheduling in the workshop.

[0011] The integration with the scheduling system is not high: the current life prediction model is usually separated from the workshop scheduling system, resulting in the prediction results cannot be efficiently applied to the optimized scheduling of actual production tasks. Summary of the Invention

[0012] To address the above issues, the present invention proposes a scheduling method for alloy processing workshops based on real-time monitoring of physical neural networks, which includes the following steps:

[0013] S1, real-time collection of machine tool physical signal data, including vibration, cutting force and temperature data;

[0014] S2, preprocessing the collected data, and obtaining enhanced time domain feature data and frequency domain feature data through Fourier transform feature extraction, which are fused as data input for the tool remaining life prediction model;

[0015] S3, the tool remaining life prediction model predicts and outputs the tool fatigue life results; the model uses a deep operator network guided by physical information, including a backbone network and a branch network to process tool wear information and physical signal data respectively, and combines the features through a fusion module;

[0016] S4, tool scheduling optimization problem modeling, using the remaining tool life prediction results of each machine tool, the processing time of each task to be processed, the life parameters of the tool consumed by the task to be processed, and the processing accuracy requirements of each task to be processed as input data, setting the objective function and constraint function; obtaining the global optimal processing scheduling plan through genetic algorithm, and then using the global optimal processing scheduling plan as the initial state, using reinforcement learning to dynamically enhance the scheduling plan, thereby obtaining the optimal scheduling plan after dynamic optimization.

[0017] Preferably, the extracted time domain feature data includes:

[0018] The mean μ of the tool wear signal reflects the overall level of the signal;

[0019] The standard deviation σ of the tool wear signal measures the volatility of the signal;

[0020] The root mean square value (RMS) of the vibration signal measures the energy level of the signal;

[0021] Peak factor τ, used to detect minor tool chipping;

[0022] The final extracted time domain feature data output is

[0023] Preferably, the frequency domain features are extracted using fast Fourier transform, and the extracted frequency domain feature data includes:

[0024] Main frequency characteristic f peak , used to monitor tool resonance. When the tool is close to failure, the main frequency will shift and the amplitude will increase;

[0025] The kurtosis SC of the cutting force signal is used to measure the energy concentration of the spectrum and reflect the energy center of the spectrum. After the tool wears, the high-frequency components increase, resulting in an increase in SC.

[0026] The energy consumption value E during the cutting process is used to measure the total energy of the signal. When the vibration and cutting force of the tool increase, the total energy E will also increase;

[0027] The output after frequency domain feature extraction is

[0028] Preferably, the backbone network is used to process tool wear information, extract high-level features related to tool life, and enhance the expression ability through nonlinear transformation, and its input is the extracted wear features The wear characteristics for:

[0029] Time domain characteristics and frequency domain characteristics Combine to form a complete feature vector The time domain features reflect the impact of tool wear on signal volatility, and the frequency domain features reflect the impact of tool wear on signal spectrum. In order to better capture the nonlinear relationship, some features are transformed to obtain enhanced features.

[0030]

[0031] Among them, log(σ+1) represents the logarithm of the standard deviation, e -τ is a negative exponential transformation, Take the square root of the center of the spectrum to make the data distribution more even;

[0032] The backbone network uses a multi-layer fully connected neural network to extract deep features:

[0033]

[0034] in, represents the weight matrix of the i-th layer, represents the bias term of the i-th layer, is the output feature of the i-th layer, and σ(·) is the ReLU activation function.

[0035] Preferably, the branch network adopts a CNN+Transformer combined architecture, and uses a CNN module to output the feature Z of the main network. T For feature extraction, the core calculation formula is:

[0036] Z C =σ(W C *Z T +b C )

[0037] Among them, W C is the convolution kernel of CNN, * represents the convolution operation, b C is the bias term, Z C is the local temporal feature extracted by CNN; σ(·) is the ReLU activation function; multi-layer convolutional CNN is used to extract features of different scales:

[0038]

[0039] The final output features of CNN are:

[0040]

[0041] Where N is the number of CNN layers, and Flatten(·) means flattening into a vector to adapt to the Transformer input;

[0042] To capture the long-term dependencies of physical signals, Transformer is used for global time series modeling:

[0043]

[0044] Among them, Q, K, V are linear transformation matrices; W Q , W K , W V is the weight matrix; Z B is the extracted time series feature; Z is obtained by multi-head attention B The attention mechanism in is calculated as:

[0045]

[0046] Among them, d kRepresents the key feature dimension; finally, Z is normalized by layer C and Z B The final model prediction output obtained after processing is:

[0047]

[0048] in, Represents the final predicted tool fatigue life.

[0049] Preferably, in order to enhance the physical consistency of the branch network, the cutting force model is used to perform physical constraint optimization:

[0050] F c =k c a p f

[0051] F t =k t a p f

[0052] Among them, F c , F t Indicates the main cutting force and feed force, k c , k t Indicates sparse cutting force, a p is the cutting depth, f is the feed rate; the physical consistency loss Lphys is defined as:

[0053] L phys =||Z B -[F c , F t ]|| 2

[0054] The final optimization target is the prediction error loss L MSE , physical consistency loss L phys and regularization loss L reg Non-linear composition:

[0055]

[0056] Among them, λ1λ2λ3 represent hyperparameters that control the relative importance of each loss term, N is the number of training samples, and Y i Indicates the actual remaining tool life. represents the predicted remaining tool life, and W represents all trainable parameters of the model.

[0057] Preferably, the tool scheduling optimization problem is modeled as follows:

[0058] First, define the variables and model the scheduling problem. Assume that there are M machine tools in the workshop, each equipped with a tool. The remaining life of each tool has been predicted and is expressed as:

[0059]

[0060] in, It is expressed as the predicted remaining life of the tool of the j-th machine tool; the workshop needs to complete N processing tasks, and each task i consumes a tool life of L i , i=1, 2, ..., N; set T i is the processing time required for task i, P i For the accuracy requirement of task i, define the scheduling decision variables:

[0061]

[0062] Among them, X ij A binary variable indicating whether task i is assigned to machine tool j. The problem goal is to find the optimal task allocation solution X = {X ij} to minimize the total cost of the task while ensuring tool life and task requirements.

[0063] Preferably, the objective function is set to optimize the cost of minimizing the overall scheduling cost, including tool replacement cost, task delay cost, and precision loss cost, as follows:

[0064]

[0065] Among them, C replace,j represents the cost of tool replacement for machine tool j, C delay，i represents the cost of delay in task i, C precision,ij represents the cost of accuracy loss due to tool wear when task i is performed by machine tool j;

[0066] The setting constraints are specifically:

[0067] Task allocation constraints: each task must be assigned to and only assigned to one machine tool;

[0068] Tool remaining life constraint: the total tool life of machine tool j cannot exceed its predicted remaining life;

[0069] Task execution time constraint: After task i is assigned to machine tool j, its deadline must be met.

[0070] Preferably, in S4, the task allocation and tool replacement strategy are optimized and solved by an adaptive genetic algorithm:

[0071] First, initialize pop chromosome codes, where each chromosome code F represents a scheduling scheme:

[0072] F=[X 11 , X 12 ,...,X NM ]

[0073] Among them, each parameter X ij The value is 0 or 1, indicating whether task i is assigned to machine tool j;

[0074] Secondly, the fitness function is calculated by the optimization target C. The smaller C is, the better the scheduling solution is:

[0075] Fitness(F)=-C(F)

[0076] Again, an adaptive roulette wheel selection strategy is used, where the selection probability is proportional to the individual fitness:

[0077]

[0078] Among them, F k is the task allocation plan for the kth individual, P(F k ) is individual F k The probability of being selected, through this step we get a new chromosome encoding population F se :

[0079]

[0080] Among them, select() represents the chromosome selection operation. After that, the present invention proposes to use adaptive two-point crossover to perform a crossover operation on the chromosome, and the crossover rate P c Dynamic Adjustment:

[0081]

[0082] Among them, t max is the maximum number of iterations, t is the current number of iterations, P c_min represents the minimum crossover probability, P c_max is the maximum crossover probability, from population F se Randomly select two chromosome individuals and As a parent, perform a crossover operation:

[0083]

[0084] Where r is a random number r~U(0,1); the new offspring population F generated after crossover cr : Afterwards, the new population enters the mutation phase;

[0085] The mutation operation is used to enhance the diversity of the population and avoid premature convergence, using an adaptive mutation method:

[0086]

[0087] Among them, P m Expressed as adaptive mutation probability, Fitness avg is the average fitness of the population, P m_min represents the minimum mutation probability, P m_max For the maximum mutation probability, the mutation process is as follows: randomly select some tasks i* and randomly adjust their assigned machine j*: F i*j =0, F i*j* =1,j*≠j; the new population individuals generated after compilation are: The solution of the new population is the global optimal scheduling solution.

[0088] Preferably, the method of dynamically enhancing the scheduling scheme by using reinforcement learning in S4 is specifically:

[0089] The new population after mutation obtained by adaptive genetic algorithm As the initial solution to be locally optimized, that is, the initial state of reinforcement learning, combined with the predicted remaining tool life And set the task queue time of machine tool j Optimize task scheduling solutions, including:

[0090] Define the state space: At time step t, the state variables of the reinforcement learning

[0091] Define the action space A t :Task redistribution is to adjust tasks to different machines: X ij ←X ik , k≠j; the tool replacement decision is to replace the tool in advance if the tool life is lower than the threshold: Task priority is adjusted to reduce the impact on tasks with high precision requirements: P i =max(P i -δ, 0);

[0092] Design a reward function and continuously optimize the scheduling plan by updating the Q value.

[0093] Compared with the prior art, the present invention has the following beneficial effects:

[0094] 1. Physical Information-Guided Deep Operator Network: This paper proposes a deep operator network that combines physical information with a deep learning model for tool life prediction. This network not only learns the dynamics of tool wear from data but also enhances the model's interpretability and predictive capabilities by incorporating physical signals (such as cutting forces, vibration, and temperature). Through physical constraint optimization, the prediction results are ensured to conform to actual physical laws, improving the model's robustness and accuracy.

[0095] 2. Deep Integration of Real-Time Monitoring and Intelligent Scheduling: By deeply integrating tool life prediction with workshop production scheduling, we propose an intelligent scheduling method based on tool remaining life. By monitoring tool status in real time and dynamically adjusting machining task allocation and tool replacement plans, we avoid overuse or premature tool replacement, optimize workshop production efficiency, and reduce production costs.

[0096] 3. Hybrid Optimization Strategy of Adaptive Genetic Algorithm and Reinforcement Learning: A hybrid optimization strategy combining an adaptive genetic algorithm and reinforcement learning is proposed to solve the shop floor task scheduling problem. This strategy uses a genetic algorithm for global optimization and reinforcement learning for local optimization. It can dynamically respond to changes in the shop floor environment (such as sudden tool life reduction and task delays), ensuring the real-time and efficient scheduling of the solution. It continuously optimizes the scheduling plan based on real-time monitoring data to ensure optimal task allocation. This reduces tool changes, optimizes tool usage, reduces maintenance costs, and improves resource utilization. BRIEF DESCRIPTION OF THE DRAWINGS

[0097] Figure 1 The figure is a flow chart of the overall method of the present invention.

[0098] Figure 2 This is a schematic diagram of the physical signal acquisition and speech processing module of the present invention.

[0099] Figure 3 This is a schematic diagram of the structure of the tool remaining life prediction model guided by physical information in the present invention.

[0100] Figure 4 A flow chart is designed for the intelligent scheduling and real-time monitoring module of the present invention.

[0101] Figure 5 This is a predicted value analysis diagram of tool life prediction using the algorithm of the present invention in an embodiment of the present invention.

[0102] Figure 6 This is a comparison chart of the scheduling costs between the algorithm of the present invention and the traditional algorithm. DETAILED DESCRIPTION

[0103] This paper proposes a scheduling method for alloy processing workshop based on real-time monitoring of physical neural network. The overall process is as follows: Figure 1 shown.

[0104] The invention will be further described below with reference to specific embodiments.

[0105] 1. Physical Signal Acquisition and Preprocessing

[0106] In the alloy processing process, tool wear and fatigue life have a vital impact on processing quality and production efficiency. In order to achieve accurate monitoring of tool status, real-time acquisition and processing of physical signal data is a key step. By installing vibration, cutting force, temperature and other sensors, multi-dimensional physical signals of the tool during the processing can be obtained. These signals provide basic data for the evaluation of tool wear status and life prediction. Figure 2 shown.

[0107] 1. Data Collection

[0108] In alloy processing workshops, the first step in real-time monitoring of tool status is to collect physical signals from machine tools. In order to accurately capture tool wear characteristics and subtle changes during the processing, a variety of high-precision sensors are used for data collection, mainly including the following types of equipment:

[0109] (1) Vibration sensor: Installed near the machine tool spindle or tool, it is used to collect the vibration signal V(t) generated by the tool during the machining process. The vibration signal can reflect the changes in the cutting force between the tool and the workpiece, and therefore has a strong indicator of the tool wear status. The sensor uses an accelerometer with a sampling frequency set to 50kHz.

[0110] (2) Cutting force sensor: Installed on the tool fixture, it can monitor the cutting force F(t) applied to the tool during machining in real time. The sensor can simultaneously measure the cutting force components in three directions: F x (t), F y (t), F z (t). Represents the cutting force of the tool in the X, Y, and Z directions, respectively. The cutting force signal is sensitive to tool wear and can effectively indicate the tool load. The sampling frequency is set to 20kHz.

[0111] (3) Temperature sensor: Since the high temperature environment during machining has a significant impact on tool wear, a temperature sensor installed in the tool cutting area is used to measure the tool's operating temperature T(t) in real time. This sensor uses an infrared thermometer for non-contact measurement, and the sampling frequency is set to 1kHz.

[0112] The analog signals of all sensors are connected to the computer system through a data acquisition card (DAQ), converted into digital signals and stored. The data collected by each sensor signal is stored in the form of a time series, so the final signal vector is:

[0113] X(t)=[V(t),F x (t), F y (t), F z (t), T(t)] (1)

[0114] Among them, X(t) is the multi-channel signal data collected by all sensors, and t is the time variable.

[0115] 2. Data Preprocessing

[0116] Since sensor signals have noise, offset and data scale differences, they must be preprocessed to improve data quality and prepare for subsequent feature extraction. The input of data preprocessing is the original signal X(t), and the processed signal X is obtained after data preprocessing. p (t).

[0117] First, filtering and noise reduction are performed to remove sensor noise and mechanical vibration interference to improve signal stability. The present invention uses a second-order Butterworth low-pass filter to remove high-frequency noise and sets the cutoff frequency f c =5kHz. The filter transfer function is:

[0118]

[0119] Among them, X fliter (t) represents the filtered signal, h(t) is the impulse response of the low-pass filter, and γ is the time variable. The data obtained by filtering and denoising based on Formula 1 is:

[0120] X fliter (t)=[V filter (t), F x，filter (t), F y，filter (t), F z，filter (t), T filter (t)] (3)

[0121] Secondly, the data is detrended and normalized so that it is mapped to the same scale range, improving the stability and convergence speed of subsequent modeling. The present invention first eliminates slow-changing trends in the signal through a linear detrending method, making the data more suitable for short-term dynamic analysis. Then, the data is standardized using the minimum-maximum normalization method, so that all input data is mapped to the range [0, 1] to eliminate the scale effect of different signals, providing high-quality input data for subsequent feature extraction and model training.

[0122] Specifically, the linear detrending Trend(X(t)) is used:

[0123] X detrend (t) = X fliter(t)-Trend(X fliter (t)) (4)

[0124] Trend(X fliter (t))=at+b (5)

[0125] Where, X detrend (t) represents the detrended signal, and a and b represent the fitting parameters calculated by the least squares method. Next, normalization is performed. Due to the different physical dimensions of cutting force, vibration, and temperature, minimum-maximum normalization is used:

[0126]

[0127] Among them, X norm (t) represents the normalized signal, and the final output data is:

[0128] X norm (t)=[V norm (t), F x，norm (t), F y，norm (t), F z，norm (t), T norm (t)] (7)

[0129] 3. Feature extraction

[0130] The health and fatigue life of a tool depend not only on the instantaneous signal value but also on the signal's overall statistical characteristics and frequency distribution. Therefore, a fast Fourier transform (FFT) is used to extract the signal's dominant frequency components and calculate the spectral energy to reflect the changes in vibration patterns during tool wear. Ultimately, all features are combined into a multidimensional feature vector, which provides input data for subsequent deep operator networks.

[0131] Specifically, the extracted features are as follows:

[0132] The mean value of the tool wear signal reflects the overall level of the signal:

[0133]

[0134] Among them, μ represents the average value of the signal, N represents the number of sampling points, X norm (i) is the normalized signal value at the i-th time step.

[0135] The standard deviation of the tool wear signal, a measure of the signal's volatility:

[0136]

[0137] Among them, σ represents the standard deviation of the signal, which reflects the stability of the tool operation.

[0138] The RMS value of the vibration signal measures the energy level of the signal. A higher RMS value indicates that the tool is subject to greater cutting forces or increased vibration:

[0139]

[0140] The peak factor is used to detect minor tool chipping. When a tool experiences localized chipping, the peak value of the vibration signal increases, causing the peak factor to increase.

[0141]

[0142] According to the above time domain feature description, the final extracted feature output is: These features can characterize the changes in the signal in the time domain during the tool wear process.

[0143] Furthermore, frequency domain feature extraction was performed to analyze the impact of tool wear on the signal spectrum. In actual machining, tool vibration signals typically contain periodic components, and the dominant frequency and energy distribution of the signal vary under different wear conditions. Therefore, a fast Fourier transform was used to extract frequency domain features to capture the spectral changes of tool wear.

[0144] Fast Fourier transform processing function:

[0145]

[0146] Among them, X FFT (f) is the spectrum of the normalized signal at frequency f, j represents the imaginary unit, and N is the number of sampling points.

[0147] Main frequency characteristic f peak Extraction: used to monitor tool resonance. When the tool is close to failure, the main frequency will shift and the amplitude will increase.

[0148]

[0149] The kurtosis SC of the cutting force signal is obtained: it is used to measure the energy concentration of the spectrum and reflect the energy center of the spectrum. After the tool wears, the high-frequency component increases, resulting in an increase in SC.

[0150]

[0151] Energy consumption during cutting: E ​​is used to measure the total energy of the signal. As tool vibration and cutting forces increase, the total energy E also increases.

[0152] E=∑ f |X FFT (f)| 2 (15)

[0153] According to the output after the above frequency domain feature extraction: These features are used to describe the spectral variation of tool status over time.

[0154] Furthermore, finally, the time domain features and frequency domain characteristics Combine to form the complete feature vector:

[0155]

[0156] Among them, the time domain features reflect the impact of tool wear on signal volatility, and the frequency domain features reflect the impact of tool wear on signal spectrum. In order to better capture the nonlinear relationship, some features are transformed to obtain enhanced features.

[0157]

[0158] Among them, log(σ+1) represents the logarithm of the standard deviation to prevent excessive changes from affecting the convergence of the model, e -τ It is a negative exponential transformation to avoid the peak factor from having too much impact on the training process. The square root of the center of the spectrum is taken to make the data distribution more uniform. The final input feature dimension of the backbone network is: Therefore, the output feature vector Use tool fatigue life prediction model.

[0159] 2. Design of tool remaining life prediction model based on physical information guidance

[0160] In the tool life prediction task, traditional machine learning methods have limitations when processing high-dimensional time series data, and it is difficult to effectively capture the dynamic changes of tool wear. Although the deep learning method can learn the time series features in the data, it lacks the physical mechanism modeling of tool wear and is easily affected by data noise. The present invention designs a deep operator network (DON) guided by physical information for tool remaining life prediction. This method can not only learn data-driven features, but also combine physical information to enhance the interpretability and predictive ability of the model. DON mainly includes a backbone network and a branch network, and through an innovative physical guidance feature fusion strategy, it achieves high-precision tool fatigue life prediction.

[0161] Deep operator network modeling Figure 3 As shown:

[0162] DON consists of a backbone network and a branch network, which process tool wear information and physical signal data respectively, and combine features through a fusion module to ultimately predict tool fatigue life.

[0163] Specifically, the backbone network is designed first. The backbone network is used to process tool wear information, extract high-level features related to tool life, and enhance the expression ability through nonlinear transformation. Its input comes from the extracted wear features. The network uses a multi-layer fully connected neural network to extract deep features:

[0164]

[0165] in, represents the weight matrix of the i-th layer, represents the bias term of the i-th layer. is the output feature of the i-th layer. σ(·) is the ReLU activation function.

[0166] Secondly, design the branch network. Since physical signal data usually has time correlation, the present invention adopts the CNN+Transformer combined architecture to realize local temporal pattern extraction and long-term dependency modeling, and enhance the model's ability to analyze dynamic signals. Specifically, the CNN module is used to output the feature Z of the backbone network. T For feature extraction, the core calculation formula is as follows:

[0167] Z C =σ(W C *Z T +b C ) (19)

[0168] Among them, W C is the convolution kernel of CNN, * represents the convolution operation, b C is the bias term, Z C is the local temporal feature extracted by CNN. σ(·) is the ReLU activation function. Multi-layer convolutional CNN is used to extract features of different scales:

[0169]

[0170] The final output features of CNN are:

[0171]

[0172] Where N is the number of CNN layers, and Flatten(·) represents flattening into a vector to adapt to the following Transformer input.

[0173] Furthermore, in order to capture the long-term dependencies of physical signals, Transformer is used for global time series modeling:

[0174]

[0175] Where Q, K, and V are linear transformation matrices. Q , W K , W V is the weight matrix. B is the extracted time series feature. Among them, Z is obtained by multi-head attention B The attention mechanism in is calculated as:

[0176]

[0177] Among them, d k Represents the key feature dimension to prevent gradient explosion. Finally, Z is normalized by layer C and Z B The final model prediction output obtained after processing is:

[0178]

[0179] in, Represents the final predicted tool fatigue life.

[0180] In addition, to enhance the physical consistency of the branch network, the cutting force model is used for physical constraint optimization:

[0181] F c =k c a p f (25)

[0182] F t =k t a p f (26)

[0183] Among them, F c , F t Indicates the main cutting force and feed force, k c , k t Indicates sparse cutting force, a p is the cutting depth, f is the feed rate. Define the physical consistency loss L phys :

[0184] L phys =||Z B -[F c , F t ]|| 2 (27)

[0185] The final optimization target is the prediction error loss L MSE , physical consistency loss L phys and regularization loss L reg Non-linear composition:

[0186]

[0187] Among them, λ1λ2λ3 represent hyperparameters that control the relative importance of each loss term, N is the number of training samples, and Y i Indicates the actual remaining tool life. represents the predicted remaining tool life, and W represents all trainable parameters of the model.

[0188] The training and optimization process of this part is as follows:

[0189] 1) Initialize parameters, set hyperparameters λ1, λ2, λ3 and learning rate;

[0190] 2) Forward propagation, calculate the output of the backbone network and branch network, calculate the fusion features, and obtain the lifespan prediction value

[0191] 3) Calculate the loss function and calculate L MSE , L phys , L reg . Calculate the total loss L.

[0192] 4) Back propagation, calculate gradients, and update network parameters.

[0193] 5) Convergence judgment: observe the change in loss and adjust the hyperparameters until the convergence conditions are met.

[0194] Through the above process, not only can the tool life be accurately predicted during the training process, It can also ensure that the prediction results conform to physical laws, while preventing the model from overfitting and ensuring generalization capabilities.

[0195] 3. Intelligent Scheduling and Real-time Monitoring Module Design

[0196] like Figure 4 As shown, the present invention is based on the predicted remaining tool life Build an intelligent scheduling model to optimize the arrangement of CNC machine tool processing tasks in the alloy processing workshop. The core goals of intelligent scheduling are:

[0197] Avoid the risk of tool failure. When the remaining life of a machine tool tool is nearing exhaustion, replace the tool in advance to prevent processing interruption or product scrapping.

[0198] Optimize task allocation and prioritize high-precision machining tasks to tools with longer lifespans to improve product quality.

[0199] Dynamically adjust scheduling strategies, combine real-time monitoring data, and promptly adjust task arrangements to ensure efficient completion of processing tasks under resource constraints.

[0200] With these goals in mind, the present invention designs and establishes an optimization model that uses a scheduling algorithm to minimize the total cost of machining tasks while ensuring that task completion time and tool life constraints are met. Based on the remaining tool life predictions and real-time monitoring data, the present invention adjusts the scheduling plan to ensure that machining tasks are executed efficiently within resource-constrained workshops.

[0201] 1. Modeling of tool scheduling optimization problem

[0202] First, define the variables and model the scheduling problem. Assume that there are M machine tools in the workshop, each equipped with a tool. The remaining life of each tool has been predicted and is expressed as:

[0203]

[0204] in, It is expressed as the predicted remaining life of the tool of the j-th machine tool. The workshop needs to complete N processing tasks, and each task i consumes tool life L i , i = 1, 2, ..., N. Set T i is the processing time required for task i, P i is the accuracy requirement of task i. Define the scheduling decision variables:

[0205]

[0206] Among them, X ij A binary variable indicating whether task i is assigned to machine tool j. The problem goal is to find the optimal task allocation solution X = {X ij} to minimize the total cost of the task while ensuring tool life and task requirements.

[0207] 2. Objective function setting

[0208] Set the objective function. The optimization goal is to minimize the comprehensive scheduling cost, including tool change cost, task delay cost, and accuracy loss cost, as follows:

[0209]

[0210] Among them, C replace，j represents the cost of tool replacement for machine tool j, C delay，i represents the cost of delay in task i, C precision,ij It represents the cost of accuracy loss due to tool wear when task i is performed by machine tool j.

[0211] Specifically, when the remaining life of the tool is not enough to complete the task, a new tool needs to be replaced, and the loss of tool replacement is set as:

[0212]

[0213] Among them, C tool The fixed cost of replacing the tool.

[0214] If task i is not completed within the specified time, task delay loss will be incurred:

[0215] C delay,i =α(T i -D i ) + (33)

[0216] Among them, D i represents the deadline of task i, and α represents the penalty factor for task delay.

[0217] High-precision tasks should be assigned to tools with less wear as much as possible, otherwise it may affect product quality, that is, loss of precision:

[0218]

[0219] Among them, β represents the weight factor of accuracy loss, and θ represents the exponential decay parameter of accuracy loss, that is, the more severe the wear, the greater the accuracy loss.

[0220] 3. Constraint setting

[0221] (1) Task allocation constraint: each task must be assigned to and only assigned to one machine tool:

[0222]

[0223] (2) Tool remaining life constraint: the total tool life of machine tool j cannot exceed its predicted remaining life:

[0224]

[0225] (3) Task execution time constraint: After task i is assigned to machine tool j, its deadline must be met:

[0226]

[0227] 4. Obtaining the global optimization plan for processing scheduling

[0228] Since the scheduling problem involves the assignment of tasks to integer decision variables X ij and continuously variable tool life The present invention adopts a hybrid solution strategy. First, the task allocation and tool replacement strategy are optimized and solved through an adaptive genetic algorithm. Then, a reinforcement learning method is used for local dynamic optimization to ensure balanced task allocation and reduce task scheduling costs.

[0229] First, initialize pop chromosome codes, where each chromosome code F represents a scheduling scheme:

[0230] F=[X 11 , X 12 ,...,X NM ] (38)

[0231] Among them, each parameter X ij The value is 0 or 1, indicating whether task i is assigned to machine tool j.

[0232] Secondly, the fitness function is calculated by the optimization target C according to formula 33 (the smaller C is, the better the scheduling solution is):

[0233] Fitness(F)=-C(F) (39)

[0234] Again, an adaptive roulette wheel selection strategy is used, where the selection probability is proportional to the individual fitness:

[0235]

[0236] Among them, F k is the task allocation plan for the kth individual, P(F k ) is individual F k The probability of being selected. Through this step, a new chromosome encoding population F is obtained. se :

[0237]

[0238] Among them, select() represents the chromosome selection operation. After that, the present invention proposes to use adaptive two-point crossover to perform a crossover operation on the chromosome, and the crossover rate P c Dynamic Adjustment:

[0239]

[0240] Among them, t max is the maximum number of iterations, t is the current number of iterations, P c_min represents the minimum crossover probability, P c_max is the maximum crossover probability. se Randomly select two chromosome individuals and As a parent, perform a crossover operation:

[0241]

[0242] Where r is a random number r~U(0,1). The new offspring population F generated after crossover cr : The new population then enters the mutation phase.

[0243] The mutation operation is used to enhance the diversity of the population and avoid premature convergence. Adaptive mutation is used:

[0244]

[0245] Among them, P m Expressed as adaptive mutation probability, Fitness avg is the average fitness of the population, P m_min represents the minimum mutation probability, P m_max is the maximum mutation probability. The mutation process is as follows: randomly select some tasks i* and randomly adjust their assigned machine j*: F i*j =0, F i*j* =1,j*≠j. The new population individuals generated after compilation are: The solution of the new population is the global optimal scheduling solution.

[0246] Afterwards, reinforcement learning is used to locally enhance the global optimal scheduling solution. Because the genetic algorithm only optimizes within a static task scheduling environment, it doesn't account for dynamic changes in the shop floor environment (such as sudden drops in tool life, task delays, and the insertion of urgent tasks). Therefore, the present invention further incorporates reinforcement learning, using a deep Q-network to locally optimize the genetic algorithm's task scheduling solution, achieving dynamic intelligent scheduling.

[0247] The new population after mutation obtained by the adaptive genetic algorithm As the initial solution to be locally optimized, that is, the initial state of reinforcement learning. Combined with the predicted remaining tool life And set the task queue time of machine tool j Optimize the task scheduling scheme. This includes the following steps:

[0248] (1) Define the state space: At time step t, the state variables of reinforcement learning

[0249] (2) Define the action space A t :Task redistribution (adjust tasks to different machines):X ij ←X ik , k≠j; the tool replacement decision is (if the tool life is lower than the threshold, replace it in advance): Task priority is adjusted to (reduce the impact on tasks with high precision requirements): P i =max(P i-δ, 0). Each action corresponds to a different task scheduling adjustment method, and the reinforcement learning model selects the optimal action through training (i.e., real-time optimization and adjustment for different local situations).

[0250] (3) Designing the reward function: the reward function R for reinforcement learning t Based on the task scheduling optimization objective, it is the same as formula (41):

[0251] R t =-C(F)(45)

[0252] The goal of reinforcement learning is to maximize the Q value and minimize the scheduling cost. By updating the Q value, the scheduling plan is continuously optimized. The Q value update formula is as follows:

[0253]

[0254] Among them, ∈ represents the learning rate, which is used to control the Q value update step size. t is the immediate reward, and the immediate optimization effect of the task scheduling solution is calculated; θ represents the discount factor, which is used to weigh the current reward against the future reward; is the maximum Q value of the next state. Q(s t , A t ) represents state s t When performing action A t The Q value after the update is the quality of the scheduling plan. After the Q value is updated, the scheduling plan is adjusted using the Q value:

[0255]

[0256] The training process of reinforcement learning based on Q learning is as follows:

[0257] ① Input the global solution to be locally optimized, that is, the mutated population

[0258] ②Calculate Q value: (s t , A t , R t , s t+1 )~D;

[0259] ③Update Q network parameters: L(ω) is the loss function:

[0260]

[0261] Among them, ω represents the parameters in the current Q network, ω - represents the target Q network parameters, which are updated periodically, η is the learning rate of the optimizer, and D is the experience replay pool, which is responsible for storing historical decision data.

[0262] ④ Update the scheduling strategy according to formula (47) to obtain the optimal task scheduling solution after local optimization.

[0263] 4. Model Deployment

[0264] To achieve intelligent workshop management, this paper deploys physical signal acquisition, deep operator networks, life prediction models, and task scheduling strategies into an integrated management platform, achieving a closed-loop management of real-time monitoring, intelligent prediction, and optimized scheduling. The specific implementation and application after deployment includes the following steps:

[0265] Physical signal acquisition: Real-time data collection of machine tool vibration, cutting force, temperature, etc. is performed, and the aforementioned data processing operations (including filtering, detrending, and Fourier transform feature extraction) are used to obtain enhanced time domain feature data and frequency domain feature data, which are then integrated as data input for the tool remaining life prediction model.

[0266] Tool Remaining Life Prediction: Using a deep operator network model based on physical signals, the remaining life of the tool is predicted and data support is provided to the task scheduling module;

[0267] Task Scheduling Optimization Module: This module involves obtaining a global optimal scheduling solution based on a genetic algorithm and locally optimizing this optimal scheduling solution through reinforcement learning to achieve dynamic task scheduling. Specifically, the module first uses ① the predicted remaining tool life of each machine tool, ② the processing time of each task to be processed, ③ the life parameters of the tool consumed by the task to be processed, and ④ the processing accuracy requirements of each task to be processed as input data to obtain a global optimal scheduling solution through a genetic algorithm. Then, using the global optimal scheduling solution as the initial state, reinforcement learning is used to dynamically enhance this scheduling solution, resulting in a dynamically optimized optimal scheduling solution. This intelligently allocates processing tasks, ensures that tasks are completed on time, and maximizes machine tool utilization.

[0268] In addition, the present invention evaluates the proposed method to verify the accuracy and rationality of the method, specifically including the following evaluation indicators.

[0269] Tool remaining life prediction accuracy evaluation: Obtain the actual tool wear data Y through the physical signal acquisition module i，j Lifespan predicted by deep operator networks To conduct an assessment:

[0270]

[0271] Among them, E j is the prediction error of the tool at machine tool j, which reflects the accuracy of the model prediction.

[0272] Furthermore, based on the scheduling cost of the final solution, the algorithm designed by the present invention is compared with the traditional GA algorithm and the fixed threshold strategy algorithm to verify the performance of the scheduling strategy based on genetic algorithm and reinforcement learning proposed by the present invention. Finally, the comparison results of the tool remaining life prediction accuracy and scheduling solution cost are shown in the attached figure. Figure 5 and 6 As shown in FIG; The results show that the method proposed in the present invention has higher prediction accuracy and lower scheduling cost.

[0273] The above description is merely a preferred embodiment of the present application and is not intended to limit the present application. Various modifications and variations are possible for those skilled in the art. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present application shall be included within the scope of protection of the present application.

[0274] Although the above describes the specific implementation methods of the present invention, it does not limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A scheduling method for alloy processing workshop based on real-time monitoring of physical neural network, characterized in that: The following processes are included: S1, real-time collection of machine tool physical signal data, including vibration, cutting force and temperature data; S2, preprocessing the collected data, and obtaining enhanced time domain feature data and frequency domain feature data through Fourier transform feature extraction, which are fused as data input for the tool remaining life prediction model; S3, the tool remaining life prediction model predicts and outputs the tool fatigue life results; the model uses a deep operator network guided by physical information, including a backbone network and a branch network to process tool wear information and physical signal data respectively, and combines the features through a fusion module; S4, tool scheduling optimization problem modeling, using the remaining tool life prediction results of each machine tool, the processing time of each task to be processed, the life parameters of the tool consumed by the task to be processed, and the processing accuracy requirements of each task to be processed as input data, setting the objective function and constraint function; obtaining the global optimal processing scheduling plan through genetic algorithm, and then using the global optimal processing scheduling plan as the initial state, using reinforcement learning to dynamically enhance the scheduling plan, thereby obtaining the optimal scheduling plan after dynamic optimization.

2. The alloy processing workshop scheduling method based on real-time monitoring of physical neural network according to claim 1, characterized in that: The extracted time domain feature data includes: The mean μ of the tool wear signal reflects the overall level of the signal; The standard deviation σ of the tool wear signal measures the volatility of the signal; The root mean square value (RMS) of the vibration signal measures the energy level of the signal; Peak factor τ, used to detect minor tool chipping; The final extracted time domain feature data output is 3. The alloy processing workshop scheduling method based on real-time monitoring of physical neural network according to claim 1, characterized in that: Fast Fourier transform is used to extract frequency domain features. The extracted frequency domain feature data includes: Main frequency characteristic f peak , used to monitor tool resonance. When the tool is close to failure, the main frequency will shift and the amplitude will increase; The kurtosis SC of the cutting force signal is used to measure the energy concentration of the spectrum and reflect the energy center of the spectrum. After the tool wears, the high-frequency components increase, resulting in an increase in SC. The energy consumption value E during the cutting process is used to measure the total energy of the signal. When the vibration and cutting force of the tool increase, the total energy E will also increase; The output after frequency domain feature extraction is 4. The method for scheduling an alloy processing workshop based on real-time monitoring of a physical neural network according to claim 1, characterized in that: The backbone network is used to process tool wear information, extract high-level features related to tool life, and enhance the expression ability through nonlinear transformation. Its input is the extracted wear features. The wear characteristics for: Time domain characteristics and frequency domain characteristics Combine to form a complete feature vector The time domain features reflect the impact of tool wear on signal volatility, and the frequency domain features reflect the impact of tool wear on signal spectrum. In order to better capture the nonlinear relationship, some features are transformed to obtain enhanced features. Among them, log(σ+1) represents the logarithm of the standard deviation, e -τ is a negative exponential transformation, Take the square root of the center of the spectrum to make the data distribution more even; The backbone network uses a multi-layer fully connected neural network to extract deep features: in, represents the weight matrix of the i-th layer, represents the bias term of the i-th layer, is the output feature of the i-th layer, and σ(·) is the ReLU activation function.

5. The method for scheduling an alloy processing workshop based on real-time monitoring of a physical neural network according to claim 1, characterized in that: The branch network adopts the CNN+Transformer combination architecture, and uses the CNN module to output the feature Z of the main network. T For feature extraction, the core calculation formula is: WITH C =σ(W C *WITH T +b C ) Among them, W C is the convolution kernel of CNN, * represents the convolution operation, b C is the bias term, Z C is the local temporal feature extracted by CNN; σ(·) is the ReLU activation function; multi-layer convolutional CNN is used to extract features of different scales: The final output features of CNN are: Where N is the number of CNN layers, and Flatten(·) means flattening into a vector to adapt to the Transformer input; To capture the long-term dependencies of physical signals, Transformer is used for global time series modeling: Among them, Q, K, V are linear transformation matrices; W Q ,W K ,W V is the weight matrix; Z B is the extracted time series feature; Z is obtained by multi-head attention B The attention mechanism in is calculated as: Among them, d k Represents the key feature dimension; finally, Z is normalized by layer C and Z B The final model prediction output obtained after processing is: in, Represents the final predicted tool fatigue life.

6. The method for scheduling an alloy processing workshop based on real-time monitoring of a physical neural network according to claim 5, characterized in that: To enhance the physical consistency of the branch network, the cutting force model is used for physical constraint optimization: F c =k c a p f F t =k t a p f Among them, F c ,F t Indicates the main cutting force and feed force, k c ,k t Indicates sparse cutting force, a p is the cutting depth, f is the feed rate; the physical consistency loss L is defined as phys : L phys =‖Z B -[F c ,F t ]‖ 2 The final optimization target is the prediction error loss L MSE , physical consistency loss L phys and regularization loss L reg Non-linear composition: Among them, λ1λ2λ3 represent hyperparameters that control the relative importance of each loss term, N is the number of training samples, and Y i Indicates the actual remaining tool life. represents the predicted remaining tool life, and W represents all trainable parameters of the model.

7. The alloy processing workshop scheduling method based on real-time monitoring of a physical neural network according to claim 1, characterized in that: The tool scheduling optimization problem is modeled as follows: First, define the variables and model the scheduling problem. Assume that there are M machine tools in the workshop, each equipped with a tool. The remaining life of each tool has been predicted and is expressed as: in, It is expressed as the predicted remaining life of the tool of the j-th machine tool; the workshop needs to complete N processing tasks, and each task i consumes a tool life of L i ,i=1,2,…,N;set T i is the processing time required for task i, P i For the accuracy requirement of task i, define the scheduling decision variables: Among them, X ij A binary variable indicating whether task i is assigned to machine tool j. The problem goal is to find the optimal task allocation solution X = {X ij } to minimize the total cost of the task while ensuring tool life and task requirements.

8. The alloy processing workshop scheduling method based on real-time monitoring of a physical neural network according to claim 7, characterized in that: The objective function is set to optimize the cost of minimizing the comprehensive scheduling cost, including tool replacement cost, task delay cost, and precision loss cost, as follows: Among them, C replace,j represents the cost of tool replacement for machine tool j, C delay,i represents the cost of delay in task i, C precision,ij represents the cost of accuracy loss due to tool wear when task i is performed by machine tool j; The constraints are set as follows: Task allocation constraints: each task must be assigned to and only assigned to one machine tool; Tool remaining life constraint, the total tool life of machine tool j cannot exceed its predicted remaining life; Task execution time constraint,After task i is assigned to machine tool j, its deadline must be met.

9. The alloy processing workshop scheduling method based on real-time monitoring of a physical neural network according to claim 1, characterized in that: In S4, the task allocation and tool replacement strategy are optimized and solved by an adaptive genetic algorithm: First, initialize pop chromosome codes, where each chromosome code F represents a scheduling scheme: F=[X 11 ,X 12 ,…,X NM ] Among them, each parameter X ij The value is 0 or 1, indicating whether task i is assigned to machine tool j; Secondly, the fitness function is calculated by the optimization target C. The smaller C is, the better the scheduling solution is: Fitness(F)=-C(F) Again, an adaptive roulette wheel selection strategy is used, where the selection probability is proportional to the individual fitness: Among them, F k is the task allocation plan for the kth individual, P(F k ) is individual F k The probability of being selected, through this step we get a new chromosome encoding population F se : Among them, select() represents the chromosome selection operation. After that, the present invention proposes to use adaptive two-point crossover to perform a crossover operation on the chromosome, and the crossover rate P c Dynamic Adjustment: Among them, t max is the maximum number of iterations, t is the current number of iterations, P c_min represents the minimum crossover probability, P c_max is the maximum crossover probability, from population F se Randomly select two chromosome individuals and As a parent, perform a crossover operation: Where r is a random number r~U(0,1); the new offspring population F generated after crossover cr : Afterwards, the new population enters the mutation phase; The mutation operation is used to enhance the diversity of the population and avoid premature convergence, using an adaptive mutation method: Among them, P m Expressed as adaptive mutation probability, Fitness avg is the average fitness of the population, P m_min represents the minimum mutation probability, P m_max For the maximum mutation probability, the mutation process is as follows: randomly select some tasks i* and randomly adjust their assigned machine j*: F i*j =0,F i*j* =1,j*≠j; the new population individuals generated after compilation are: The solution of the new population is the global optimal scheduling solution.

10. The alloy processing workshop scheduling method based on real-time monitoring of a physical neural network according to claim 7, characterized in that: The specific method of using reinforcement learning to dynamically enhance the scheduling scheme in S4 is as follows: The new population after mutation obtained by adaptive genetic algorithm As the initial solution to be locally optimized, that is, the initial state of reinforcement learning, combined with the predicted remaining tool life And set the task queue time of machine tool j Optimize task scheduling solutions, including: Define the state space: At time step t, the state variables of the reinforcement learning Define the action space A t :Task redistribution is to adjust tasks to different machines: X ij ←X ik ,k≠j; the tool replacement decision is to replace the tool in advance if the tool life is lower than the threshold: Task priority is adjusted to reduce the impact on tasks with high precision requirements: P i =max(P i -δ,0); Design a reward function and continuously optimize the scheduling plan by updating the Q value.

Images