Intelligent production scheduling and early warning method and system for hardware processing

Abstract

This invention relates to the field of intelligent production scheduling and early warning methods, and more specifically, to an intelligent production scheduling and early warning method and system for hardware processing. The method includes the following steps: acquiring real-time production scheduling operation data on the hardware processing production line; preprocessing the real-time production scheduling operation data to obtain preprocessed data; dividing the preprocessed data into dynamic time windows; extracting multi-scale features from the divided time window data; using a quantum-inspired deep autoencoder network to perform anomaly pattern recognition on the extracted multi-scale features; diagnosing the causes of anomalies based on the anomaly pattern recognition results; generating and pushing early warning information; and updating the model online based on feedback information. Through the innovative quantum-inspired deep autoencoder network, complex anomaly patterns can be accurately identified, significantly improving the accuracy, precision, and recall of anomaly detection. Test results show that the accuracy reaches 98.7%, far exceeding traditional methods.

Description

Technical Field

[0001] This invention relates to the field of intelligent production scheduling and early warning methods, and more specifically, to an intelligent production scheduling and early warning method and system for metal processing. Background Technology

[0002] With the rapid development of Industry 4.0 and intelligent manufacturing, the hardware processing industry is facing unprecedented opportunities and challenges. Efficient and accurate production scheduling and real-time early warning systems have become key factors in enhancing enterprise competitiveness. However, existing scheduling and early warning technologies still have many shortcomings when facing complex and ever-changing hardware processing environments.

[0003] Traditional rule-based early warning methods were once the mainstream in the industry. This approach relies on expert experience to establish fixed thresholds and rules. While simple to implement, its limitations are becoming increasingly apparent in the face of the complexities of modern metal processing. First, fixed thresholds are difficult to adapt to the needs of different working conditions and product types, leading to a high false alarm rate. Second, rule-based systems struggle to capture complex multivariate relationships and cannot identify subtle abnormal patterns. Third, this method lacks adaptability and cannot be updated to keep pace with changes in the production environment.

[0004] In recent years, machine learning methods, such as random forests and support vector machines, have begun to be applied to the field of production scheduling early warning. These methods have improved the accuracy and flexibility of early warning to some extent. However, they still have some problems: First, these methods usually require a large amount of labeled data for training, while in actual production, abnormal samples are often scarce; second, their ability to process high-dimensional, time-series data is limited, making it difficult to fully utilize the rich information in the metal processing process; finally, these methods have poor interpretability, making it difficult to provide in-depth root cause analysis for anomalies.

[0005] With the development of deep learning technology, some studies have begun to attempt to use methods such as LSTM autoencoders for anomaly detection. These methods perform well in processing time-series data and can capture more complex patterns. However, they still face some challenges: First, deep learning models typically require a large amount of computational resources, making it difficult to achieve real-time alerts; second, these models are prone to overfitting and lack robustness when faced with new anomaly types; third, the "black box" nature of deep learning models makes the results difficult to interpret, hindering engineers from subsequent problem diagnosis and handling.

[0006] Furthermore, existing methods generally share a common problem: they often separate anomaly detection, diagnosis, and early warning, lacking a unified framework to integrate these key aspects. This leads to information fragmentation, affecting the overall effectiveness of the early warning system.

[0007] Given these shortcomings of existing technologies, there is an urgent need for an intelligent scheduling and early warning method and system for hardware processing. Summary of the Invention

[0008] In response to the above problems, an intelligent production scheduling and early warning method and system for hardware processing were proposed.

[0009] This invention provides an intelligent production scheduling and early warning method for hardware processing, comprising the following steps:

[0010] The system acquires real-time production scheduling data from the hardware processing production line; preprocesses the real-time production scheduling data to obtain preprocessed data; dynamically divides the time window based on the preprocessed data; extracts multi-scale features from the divided time window data; uses a quantum-inspired deep autoencoder network to identify abnormal patterns in the extracted multi-scale features; diagnoses the causes of abnormalities based on the abnormal pattern identification results; generates and pushes early warning information; and updates the model online based on feedback information.

[0011] Preferably, the preprocessing step includes:

[0012] An improved Hampel filter is used to remove outliers, where the filter is calculated using the following formula:

[0013]

[0014] Where, x′ i For the filtered data points, x i For the original data points, X w For x i The sliding window is centered, MAD is the absolute deviation of the median, and γ is an adjustable parameter.

[0015] Preferably, the dynamic time window division step includes:

[0016] The formula for calculating the adaptive window size W is as follows:

[0017]

[0018] Where ACF is the autocorrelation function and threshold is the preset threshold; an overlapping sliding window sequence is generated based on the adaptive window size W and the preset overlap rate r.

[0019] Preferably, the multi-scale feature extraction step includes:

[0020] The formula for calculating the sample entropy SE(τ) at multiple time scales τ is as follows:

[0021] Where A m(r) represents the probability of template matching in the m-dimensional phase space under tolerance r;

[0022] Multi-scale frequency domain features (WPT) obtained by applying wavelet packet transform i The calculation formula is as follows:

[0023] WPT i =WPT(S i ,level=3,wavelet='db4')

[0024] Where S i The input signal is denoted by 'level', the decomposition level is denoted by 'level', and the wavelet is denoted by 'wavelet', representing the wavelet basis functions.

[0025] Morphological features MG(S) are extracted using the morphological gradient operator. i The calculation formula is as follows:

[0026]

[0027] Among them ⊕ and These represent expansion and corrosion operations, respectively, with B being the structural element.

[0028] Preferably, the quantum-inspired deep autoencoder network comprises:

[0029] Quantum state attention mechanism, where attention weight w i The calculation formula is:

[0030] w i =|α i | 2 =|<ψ|U i |ψ>| 2

[0031] Where |ψ> represents the quantum state, |i> represents the computational ground state, and α i For complex amplitude, U i For quantum gate operations; fractal convolution module, where the fractal convolution kernel F k The formula for generating it is:

[0032] F k =IFS(W,k,s)

[0033] Where IFS is the iterative function system, W is the set of transformation matrices, k is the number of iterations, and s is the initial shape.

[0034] Topology adaptive connectivity structure, where network connectivity strength C ij The calculation formula is:

[0035] C ij =σ(W·Φ(X) i ,Xj )+b)

[0036] Where Φ is the continuous cohomology function, σ is the activation function, and W and b are learnable parameters.

[0037] The non-equilibrium quantum annealing optimization module, where parameter updates follow the non-equilibrium quantum dynamics equations:

[0038]

[0039] Where ρ is the density matrix, H is the Hamiltonian, and L is the dissipative superoperator.

[0040] Preferably, the abnormal pattern recognition step further includes:

[0041] The abnormality score AS is calculated using the following formula:

[0042] AS = log(RE) + λ·D W (X,X ref )

[0043] Where RE is the reconstruction error, and D W Let X be the Wasserstein distance. ref As a reference to normal samples, λ is a balance factor;

[0044] Based on adaptive threshold T adaptive Anomalies are identified, where T adaptive The update formula is:

[0045] T adaptive =β·T adaptive +(1-β)·(μ(AS)+γ·σ(AS))

[0046] Where μ and σ are the mean and standard deviation functions, respectively, β is the smoothing factor, and γ is the sensitivity parameter.

[0047] Preferably, the abnormality cause diagnosis step includes:

[0048] Feature importance φ is calculated using the Shapley value-based SHAP method. i The calculation formula is as follows:

[0049]

[0050] Where F is the feature set, and v is the value function of the feature subset;

[0051] A causal graph G(V,E) is constructed based on the PC algorithm, where V is the set of nodes and E is the set of edges. A dynamic Bayesian network is used to simulate the propagation of anomalies in the causal graph, and its conditional probability distribution is as follows:

[0052] P(Xt |Pa(X t ))=f(X t ,Pa(X t ),θ)

[0053] Where Pa(X) t ) is X t The set of parent nodes, f is the conditional probability distribution function, and θ is the model parameter; calculate the parent node X for each node. i Root cause score RCS(X) i The calculation formula is as follows:

[0054]

[0055] Where P(path|G,AS) is the probability that an anomaly propagates through path path given an anomaly score AS.

[0056] Preferably, the step of generating and pushing early warning information includes:

[0057] The Comprehensive Severity Index (SDI) is calculated using the following formula:

[0058] SDI=w1·normalize(AS)+w2·max(normalize(RCS))

[0059] Where w1 and w2 are weighting factors, normalize is the normalization function, AS is the anomaly score, and RCS is the root cause score.

[0060] Generate JSON-formatted alert information that includes timestamps, anomaly scores, severity indices, affected components, root causes and their importance and causal probability, and recommended actions;

[0061] Early warning information is pushed based on a multi-level push strategy, including:

[0062] When SDI <T low At that time, only log entries are recorded;

[0063] When T low ≤SDI <T high At that time, it is pushed to the monitoring system;

[0064] When SDI≥T high In such cases, information is immediately pushed to relevant personnel through multiple channels.

[0065] Preferably, the online model update step includes:

[0066] Incremental learning is performed using an elastic weight merging algorithm based on the Fisher information matrix, with the loss function L(θ) as follows:

[0067]

[0068] Where L new For the loss function of the new data, F i θ represents the diagonal elements of the Fisher information matrix. * The ADWIN algorithm is used to detect concept drift for the old model parameters. For any partition (W0, W1) of the sliding window W, if the following conditions are met:

[0069] |μ(W0)-μ(W1)|>∈ cut

[0070] Concept drift was detected, where m is the window size, and δ is the confidence level parameter;

[0071] Based on the concept drift detection results, neural architecture search technology is used to dynamically adjust the model structure, with the following optimization objective:

[0072]

[0073] Where a represents the architecture, w represents the weights, R represents the reward function, and π represents the architecture distribution strategy. An improved knowledge distillation technique is applied to transfer knowledge from the old model to the new model, with the loss function L... KD for:

[0074] L KD =α·L CE (y,σ(z s / T))+(1-α)·L KL (α(z t / T),σ(z s / T))

[0075] Where L CE For cross-entropy loss, L KL Let z be the KL divergence. t and z s α represents the logits for the teacher and student models, respectively, where T is the temperature parameter and α is the equilibrium factor.

[0076] A hardware processing intelligent scheduling and early warning system for implementing the method includes:

[0077] The data acquisition module is used to acquire real-time production scheduling data on the hardware processing production line;

[0078] The preprocessing module is used to preprocess the real-time production scheduling data;

[0079] The time window segmentation module is used to perform the dynamic time window segmentation;

[0080] The feature extraction module is used to perform the multi-scale feature extraction.

[0081] Anomaly detection module, used to perform anomaly pattern detection using the quantum-inspired deep autoencoder network;

[0082] The diagnostic module is used to diagnose the cause of the abnormality.

[0083] An early warning module is used to generate and push the early warning information; and

[0084] The update module is used to update the model online.

[0085] The main beneficial effects of this invention include:

[0086] 1. High Accuracy: Through an innovative quantum-inspired deep autoencoder network (QDAN), this invention can accurately identify complex anomaly patterns, significantly improving the accuracy, precision, and recall of anomaly detection. Test results show that the accuracy of this invention reaches 98.7%, far exceeding that of traditional methods.

[0087] 2. Rapid Response: Thanks to the parallel computing capabilities of QDAN and the unbalanced quantum annealing optimization algorithm, this invention can complete anomaly detection in an extremely short time. The average detection time is only 12.3 seconds, significantly better than other methods, providing a possibility for timely prevention of production problems.

[0088] 3. Low false alarm rate: The false alarm rate of this invention is only 1.2%, which greatly reduces unnecessary production interruptions and improves the reliability and production efficiency of the system.

[0089] 4. In-depth root cause analysis: By integrating the SHAP method, causal graph construction, and dynamic Bayesian network diagnostic modules, this invention can provide accurate root cause analysis with an accuracy rate of 93.8%. This provides strong support for engineers to quickly solve problems.

[0090] 5. Adaptability: Through the online update module, including incremental learning, concept drift detection and model structure adaptation, this invention can continuously learn and optimize, adapt to the dynamically changing production environment, and maintain long-term high performance.

[0091] 6. Multi-scale analysis: By employing multi-scale feature extraction technology, this invention can capture abnormal patterns at different time and frequency scales, thereby improving the comprehensiveness and robustness of the system.

[0092] 7. Overall Framework: This invention provides a unified framework that organically combines anomaly detection, diagnosis, and early warning, realizing intelligent management of the entire process from data collection to early warning push.

[0093] 8. Interpretability: Through quantum state attention mechanism and topological adaptive connection structure, this invention provides interpretability of model decision-making while maintaining high performance, which helps users understand and trust the output of the system.

[0094] In summary, this invention not only solves the problems of insufficient accuracy, slow response speed, high false alarm rate, and lack of adaptive capability in existing technologies, but also provides a comprehensive, efficient, and reliable intelligent scheduling and early warning solution for hardware processing. This will significantly improve the production efficiency, product quality, and equipment utilization rate of hardware processing enterprises, providing strong support for the intelligent transformation of the industry. Attached Figure Description

[0095] Figure 1 This is a flowchart of an intelligent scheduling and early warning method for hardware processing according to the present invention.

[0096] Figure 2 This is a schematic diagram of the framework of an intelligent production scheduling and early warning system for hardware processing according to the present invention. Detailed Implementation

[0097] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of an intelligent scheduling and early warning method for hardware processing proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0098] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0099] The technical solution of the present invention will be clearly and completely described below with reference to specific embodiments.

[0100] In modern hardware processing plants, various devices on the production line continuously generate a large amount of real-time data. This data includes key information such as equipment operating status, production progress, and product quality. The intelligent production scheduling and early warning method and system proposed in this invention aim to fully utilize this data to promptly detect potential production anomalies and provide accurate diagnosis and early warning, thereby improving production efficiency and reducing the risk of failure.

[0101] like Figure 1-2As shown, the intelligent production scheduling and early warning system for hardware processing of the present invention includes a data acquisition module 1, a preprocessing module 2, a time window division module 3, a feature extraction module 4, an anomaly identification module 5, a diagnosis module 6, an early warning module 7, and an update module 8. These modules work together to realize the complete process from data acquisition to early warning push.

[0102] Specifically, the intelligent scheduling and early warning method for hardware processing of the present invention includes the following steps:

[0103] 1. Data Acquisition: Data acquisition module 1 is responsible for acquiring real-time production scheduling data from the hardware processing production line. In a typical hardware processing workshop, we might collect the following types of data:

[0104] Equipment parameters: such as spindle speed, feed rate, and tool temperature of CNC machine tools.

[0105] Production progress: such as processing time for each workpiece, completion rate of the current batch, etc.

[0106] Quality indicators: such as workpiece dimensional deviation, surface roughness, etc.

[0107] These data are typically collected by sensors distributed throughout the production line, such as the ABB FieldSensors series of sensors. The frequency of data acquisition may vary depending on the parameter type, typically ranging from 0.1 seconds to 1 minute.

[0108] 2. Data preprocessing: Preprocessing module 2 cleans and standardizes the collected raw data. This step is crucial because the raw data may contain noise, outliers, or missing values, all of which can affect the accuracy of subsequent analysis.

[0109] An improved Hampel filter is used to remove outliers, where the filter is calculated using the following formula:

[0110]

[0111] Where, x′ i For the filtered data points, x i For the original data points, X w For x i The sliding window is centered, MAD is the absolute deviation of the median, and γ is an adjustable parameter, usually set to 3.

[0112] For example, suppose we are processing spindle speed data for a CNC machine tool. Under normal circumstances, the speed may fluctuate between 5000-6000 RPM. If a reading of 9000 RPM suddenly appears, the Hampel filter will identify it as an outlier and replace it with the median of the surrounding data.

[0113] 3. Dynamic Time Window Segmentation: Module 3 is responsible for dividing the continuous data stream into a series of overlapping time windows. The purpose of this step is to capture the temporal correlation of the data and provide a foundation for subsequent feature extraction.

[0114] This invention employs an adaptive window size method, the calculation formula of which is:

[0115]

[0116] Where ACF is the autocorrelation function, and threshold is a preset threshold, usually set to 0.2. An overlapping sliding window sequence is generated based on the adaptive window size W and the preset overlap rate r.

[0117] For example, if spindle speed data shows a periodic change approximately every 100 seconds, the window size might be set to approximately 100 seconds.

[0118] The system will then generate an overlapping sliding window sequence using a preset overlap rate r (usually 0.5):

[0119] S i =[F t ,F t+1 ,...,F t+W-1 ],t=i×W×(1-r)

[0120] This results in 50% data overlap between adjacent windows, which helps capture patterns across windows.

[0121] 4. Multi-scale feature extraction: Feature extraction module 4 performs multi-scale feature extraction on the data for each time window. This step aims to extract features from the raw data that can effectively characterize the system state.

[0122] This invention employs three feature extraction methods.

[0123] 4.1 Sample Entropy: Calculate the sample entropy SE(τ) at multiple time scales τ. The formula is as follows:

[0124]

[0125] Among them, A m (r) is the probability of template matching in the m-dimensional phase space under tolerance r; τ is usually taken as [1,5,10]; m is usually taken as 2; r is usually taken as 0.2 times the standard deviation;

[0126] Sample entropy can reflect the complexity of a time series. For example, for spindle speed data, a sudden increase in sample entropy may indicate that the equipment is becoming unstable.

[0127] 4.2 Wavelet Packet Transform: Applying wavelet packet transform to obtain multi-scale frequency domain features:

[0128] WPT i =WPT(S i ,level=3,wavelet='db4')

[0129] Among them, S i This is the input signal, level is the decomposition level (3 here), and wavelet is the wavelet basis function (Daubechies4 wavelet is used here). Wavelet packet transform can provide information on the energy distribution of the signal in different frequency bands. For example, if the energy in a certain frequency band suddenly increases, it may mean that the equipment is experiencing abnormal vibration.

[0130] 4.3 Morphological gradient: Extracting morphological features using morphological gradient operators.

[0131]

[0132] Among them ⊕ and These represent expansion and corrosion operations, respectively, with B being the structural element.

[0133] Morphological gradients can detect rapid changes in signals. For example, if there is a sudden jump in the dimensional deviation data of a workpiece, the morphological gradient will show a significant response.

[0134] 5. Anomaly Pattern Recognition: The anomaly recognition module 5 is the core of this invention. It uses an innovative quantum-inspired deep autoencoder network (QDAN) to identify anomaly patterns. This network combines concepts from quantum computing, fractal geometry, and topology, enabling it to effectively capture complex anomaly patterns.

[0135] 5.1 Quantum State Attention Mechanism: First, QDAN uses the quantum state attention mechanism to assign weights to different features. Attention weight w i The calculation formula is:

[0136] w i =|α i | 2 =|<ψ|U i |ψ>| 2

[0137] Where |ψ> is the quantum state; |i> is the computational ground state; α i It is a complex amplitude; U i It is a quantum gate operation;

[0138] In practical applications, we can map different characteristics (such as spindle speed, feed rate, etc.) to different quantum ground states. Quantum gate operation U iThe system dynamically adjusts its attention weight based on the importance of the features. For example, if the spindle speed is found to be particularly important for anomaly detection, the system will automatically increase its corresponding attention weight.

[0139] 5.2 Fractal Convolution Module. Next, QDAN uses the fractal convolution module to process multi-scale features. The fractal convolution kernel F... k The formula for generating it is:

[0140] F k =IFS(W,k,s)

[0141] IFS is an iterative function system; W is a set of transformation matrices; k is the number of iterations, usually 3-5; s is the initial shape, which can be a simple geometric shape such as a triangle.

[0142] Fractal convolution can capture self-similarity in data. For example, during processing, certain anomalous patterns may exhibit similar structures at different time scales, and fractal convolution is particularly suitable for capturing such patterns. Output feature map G k The calculation formula is:

[0143] G k =σ(F k *X+b k )

[0144] Where σ is the activation function, here GELU is used, * denotes convolution operation, X is the input feature, and b k It is a bias term.

[0145] 5.3 Topology Adaptive Connection Structure QDAN also introduces an adaptive connection structure based on algebraic topology. First, define the persistent homology function Φ:

[0146] Φ(X)={H k (X ε )|ε∈R + ,k∈Z +}

[0147] Among them, H k It is a k-order homology group; X ε It is an ε-neighborhood complex of X;

[0148] Network connection strength C i j is dynamically adjusted based on the continuous homology characteristic:

[0149] C i j=σ(W·Φ(X i ,X j )+b)

[0150] Here, W and b are learnable parameters, and this structure can adaptively capture the topological features of the data. For example, if certain anomalous patterns form specific topological structures (such as loops or holes) in the feature space, this connection method can better identify these patterns.

[0151] 5.4 Non-equilibrium Quantum Annealing Optimization QDAN uses the non-equilibrium quantum annealing algorithm to optimize network parameters. First, define the Hamiltonian H:

[0152]

[0153] Where, σ z and σ x It is a Pauli matrix; J ij Γ is the spin coupling strength; Г is the transverse field strength.

[0154] Parameter updates follow nonequilibrium quantum dynamics:

[0155] Where ρ is the density matrix and L is the dissipative superoperator, this optimization method can effectively avoid getting trapped in local optima and improve the generalization ability of the model.

[0156] 5.5 Calculation of Outlier Scores: QDAN calculates outlier scores AS.

[0157] AS = log(RE) + λ·D W (X,X ref )

[0158] Where RE is the reconstruction error, and D W It is the Wasserstein distance, X ref This is a reference normal sample; λ is a balance factor, typically ranging from 0.5 to 1.0.

[0159] 5.6 The adaptive threshold system uses an adaptive threshold T adaptive To determine anomalies:

[0160] T adaptive =βT adaptive +(1-β)·(μ(AS)+γ·σ(AS))

[0161] Where μ and σ are the mean and standard deviation functions, respectively, β is the smoothing factor, usually taken as 0.9-0.99, and γ is the sensitivity parameter, usually taken as 2-3. If AS>T a The daptive option is considered abnormal.

[0162] 6. Abnormal Cause Diagnosis: Diagnosis module 6 is responsible for analyzing the detected abnormalities and identifying possible causes.

[0163] 6.1 Feature Importance Calculation: The feature importance φ is calculated using the SHAP method based on Shapley values. i :

[0164]

[0165] Where F is the feature set and v is the value function of the feature subset; for example, if the SHAP value shows that the spindle speed contributes the most to a certain anomaly, then there may be a problem with the spindle.

[0166] 6.2 Causal Graph Construction A causal graph G(V,E) is constructed using the PC algorithm, where V is the set of nodes and E is the set of edges. This helps in understanding the causal relationships between different parameters.

[0167] 6.3 Anomaly Propagation Analysis: Using Dynamic Bayesian Networks to Simulate Anomaly Propagation in the Cause-and-Effect Graph.

[0168] P(X t |Pa(X t ))=f(X t ,Pa(X t ),θ)

[0169] Wherein, Pa(X) t ) is X t The set of parent nodes; f is the conditional probability distribution function; θ is the model parameters.

[0170] 6.4 Root cause score calculation: Calculate X for each node. i Root cause score RCS(X) i ):

[0171]

[0172] Where P(path|G,AS) is the probability that an anomaly propagates through path path given an anomaly score AS.

[0173] A causal graph G(V,E) is constructed based on the PC algorithm, where V is the set of nodes and E is the set of edges. A dynamic Bayesian network is used to simulate the propagation of anomalies in the causal graph, and its conditional probability distribution is as follows:

[0174] P(X t |Pa(X t ))=f(X t ,Pa(X t ),θ)

[0175] Where Pa(X) t ) is X t The set of parent nodes, f is the conditional probability distribution function, and θ is the model parameter; calculate the parent node X for each node. i Root cause score RCS(X)i The calculation formula is as follows:

[0176]

[0177] Where P(path|G,AS) is the probability that an anomaly propagates through path path given an anomaly score AS.

[0178] 7. Early warning information generation and push: Early warning module 7 is responsible for generating early warning information and pushing it according to the severity.

[0179] 7.1 Calculation of Comprehensive Severity Index

[0180] Calculate the Comprehensive Severity Index (SDI):

[0181] SDI=w1·normalize(AS)+w2·max(normalize(RCS))

[0182] Where w1 and w2 are weighting factors, typically w1 = 0.6 and w2 = 0.4; normalize is the normalization function that maps the values ​​to the [0,1] interval; AS is the anomaly score and RCS is the root cause score.

[0183] 7.2 The warning information is structured. The system generates warning information in JSON format, which includes the following fields.

[0184] timestamp: Current timestamp; anomaly_score: Anomaly score (AS); severity_index: Overall severity index (SDI); affected_components: List of affected components; root_causes: Feature name and importance for each possible root cause. And causal probability; recommended_actions: a list of recommended actions.

[0185] 7.3 Multi-level push strategy: Based on SDI value, the system adopts a multi-level push strategy.

[0186] When SDI <T low When the value is 0.3, only log entries are recorded.

[0187] When T low ≤SDI <T high When the value is 0.7, it is pushed to the monitoring system.

[0188] When SDI≥T high In such cases, information is immediately pushed to relevant personnel through multiple channels (SMS, email, application notifications).

[0189] 8. Online model updates: Module 8 is responsible for continuously optimizing the model to adapt to dynamic changes in the production environment.

[0190] 8.1 Incremental learning is performed using the Elastic Weights Combined (EWC) algorithm based on the Fisher information matrix. The loss function L(θ) is:

[0191]

[0192] Where L new It is the loss function for the new data; F i These are the diagonal elements of the Fisher information matrix; θ * These are the parameters of the old model; λ is the regularization coefficient, typically ranging from 0.1 to 18.2.

[0193] 8.2 Concept drift detection: The ADWIN (Adaptive Windowing) algorithm is used to detect changes in data distribution. For any partition (W0, W1) of the sliding window W, if the following conditions are met:

[0194] |μ(W0)-μ(W1)|>ε cut

[0195] Concept drift was detected. Among them:

[0196] m is the window size; δ is the confidence level parameter, usually taken as 0.05; n is the total number of samples.

[0197] 8.3 Adaptive Model Structure: Based on concept drift detection results, the model structure is dynamically adjusted using Neural Architecture Search (NAS) technology. The optimization objective is:

[0198]

[0199] Where a is the architecture, w is the weight, R is the reward function, and π is the architecture distribution strategy.

[0200] 8.4 Knowledge Distillation: This section describes the application of improved knowledge distillation techniques to transfer knowledge from the old model to the new model. The loss function L... KD for:

[0201]

[0202] Among them, L CE It is the cross-entropy loss, L KL It is the KL divergence, z t and z S These are the logits of the teacher and student models, respectively. T is the temperature parameter, usually taken as 2-5, and α is the balance factor, usually taken as 0.5-0.7.

[0203] The intelligent scheduling and early warning system for hardware processing of this invention is a complex system integrating multiple functional modules, designed to achieve real-time monitoring, anomaly detection, and early warning of hardware processing production lines. The core components of the system include:

[0204] Data Acquisition Module 1: This module is responsible for real-time data acquisition from various sensors and equipment on the hardware processing production line. It uses a high-precision sensor network (such as the ABB FieldSensors series) to collect data on equipment parameters, production progress, and quality indicators. The data acquisition frequency is dynamically adjusted according to different parameter types to ensure that key information is captured without generating excessive redundant data.

[0205] Preprocessing Module 2: Preprocessing Module 2 receives the raw data and performs cleaning and standardization. It uses an improved Hampel filter to remove outliers, ensuring data quality for subsequent analysis. This module is also responsible for handling missing values, potentially using methods such as interpolation or forward imputation.

[0206] Time window segmentation module 3: This module divides the continuous data stream into a series of overlapping time windows. It uses an adaptive window size method, dynamically adjusting the window size based on the autocorrelation of the data to capture the temporal correlation of the data.

[0207] Feature Extraction Module 4: Feature extraction module 4 performs multi-scale feature extraction on the data for each time window. It comprehensively uses methods such as sample entropy, wavelet packet transform, and morphological gradient to extract features that can characterize the system state from different perspectives.

[0208] Anomaly Detection Module 5: This is the core module of the system, employing an innovative quantum-inspired deep autoencoder network (QDAN) for anomaly pattern recognition. QDAN combines concepts from quantum computing, fractal geometry, and topology, effectively capturing complex anomaly patterns. This module also includes an adaptive threshold mechanism that dynamically adjusts anomaly judgment criteria based on the current production status.

[0209] Diagnostic Module 6: Diagnostic Module 6 is responsible for analyzing detected anomalies and identifying possible causes. It uses the SHAP method to calculate feature importance, constructs a cause-effect graph, and uses a dynamic Bayesian network to simulate anomaly propagation, ultimately calculating the root cause score for each possible cause.

[0210] Early Warning Module 7: Early Warning Module 7 is responsible for generating early warning information and pushing it out according to its severity. It calculates a comprehensive severity index, generates structured early warning information, and sends the information to relevant personnel or systems according to a multi-level push strategy.

[0211] Update Module 8: Update Module 8 is responsible for continuously optimizing the model to adapt to dynamic changes in the production environment. It uses techniques such as incremental learning, concept drift detection, model structure adaptation, and knowledge distillation to ensure that the system remains efficient and accurate over time. These modules work closely together to form a closed-loop intelligent early warning system.

[0212] To verify the superiority of the present invention, we designed one embodiment and three comparative examples, and conducted detailed comparative experiments.

[0213] Example 1: The intelligent scheduling and early warning method for hardware processing according to the present invention

[0214] Comparative Example 1: Traditional Rule-Based Early Warning Methods

[0215] Comparative Example 2: Simple Machine Learning Methods (Random Forest)

[0216] Comparative Example 3: Conventional Deep Learning Methods (LSTM Autoencoder)

[0217] Testing Environment: We conducted a three-month test at a medium-sized hardware processing plant. The plant has 20 CNC machine tools, operating 24 hours a day. We collected multi-dimensional data including spindle speed, feed rate, tool temperature, machining time, and product dimensional deviations.

[0218] Test metrics:

[0219] 1. Anomaly detection accuracy: The ratio of the number of correctly identified samples to the total number of samples.

[0220] 2. Precision: The ratio of the number of correctly identified anomalous samples to the total number of samples identified as anomalous.

[0221] 3. Recall: The ratio of the number of correctly identified anomalous samples to the total number of actual anomalous samples.

[0222] 4. F1 score: The harmonic mean of precision and recall.

[0223] 5. Average Detection Time (ADT): The average time from when an anomaly occurs to when it is detected by the system.

[0224] 6. False Alarm Rate (FAR): The proportion of normal samples that are incorrectly identified as abnormal.

[0225] 7. Root Cause Accuracy (RCA): The percentage of cases in which the root cause of an anomaly is correctly identified.

[0226] We used cross-validation, splitting the dataset into a training set (70%) and a test set (30%). For each method, we performed 10 cross-validations and took the average as the final result.

[0227] The test results are shown in the table below:

[0228] index Example 1 Comparative Example 1 Comparative Example 2 Comparative Example 3 Accuracy (%) 98.7 85.3 92.1 95.4 Accuracy (%) 97.5 78.6 89.3 93.2 Recall rate (%) 96.9 72.4 86.7 91.8 F1 score 0.972 0.754 0.880 0.925 Average detection time (s) 12.3 87.5 45.2 28.7 False alarm rate (%) 1.2 8.7 4.5 2.8 Root cause analysis accuracy (%) 93.8 61.2 78.5 85.1

[0229] Based on the test results, the intelligent scheduling and early warning method for hardware processing of the present invention (Example 1) is significantly better than other comparative methods in all indicators, and therefore can be considered the best embodiment.

[0230] The results are analyzed and interpreted as follows:

[0231] 1. Anomaly Detection Performance: This invention achieves state-of-the-art results in accuracy, precision, recall, and F1 score. This is primarily attributed to the innovative design of the Quantum-Inspired Deep Autoencoder Network (QDAN). The quantum-state attention mechanism effectively captures complex relationships between features, the fractal convolution module can identify multi-scale anomaly patterns, and the topologically adaptive connection structure enhances the model's adaptability to changes in data structure.

[0232] 2. Detection Speed: The average detection time of this invention is only 12.3 seconds, far faster than other methods. This rapid response capability stems from the parallel computing characteristics of QDAN and the unbalanced quantum annealing optimization algorithm, enabling the model to complete the identification of abnormal patterns in a very short time.

[0233] 3. False Alarm Rate: The false alarm rate of this invention is only 1.2%, significantly lower than other methods. This demonstrates that QDAN can not only accurately identify anomalies but also effectively distinguish between normal fluctuations and genuine anomalies, which is crucial for reducing unnecessary production interruptions.

[0234] 4. Root Cause Analysis Accuracy: This invention also significantly outperforms other methods in root cause analysis accuracy, reaching 93.8%. This is mainly due to the integrated SHAP method, causal graph construction, and dynamic Bayesian network, which together constitute a powerful anomaly diagnosis framework.

[0235] 5. Adaptability and Robustness: Although not directly reflected in the table, during the three-month testing period, we observed that the method of this invention can adapt well to the dynamic changes in the production environment. This is attributed to the system's online update module, which includes techniques such as incremental learning, concept drift detection, and model structure adaptation.

[0236] The intelligent scheduling and early warning method for hardware processing of this invention significantly outperforms existing methods in terms of accuracy, speed, reliability, and interpretability in anomaly detection. It not only quickly and accurately identifies anomalies but also provides in-depth root cause analysis while maintaining a low false alarm rate. These advantages enable this method to provide a more reliable and efficient intelligent early warning solution for the hardware processing industry, helping to improve production efficiency, reduce equipment failures and quality problems, thereby creating significant economic value for enterprises.

[0237] Furthermore, the adaptive capability of the method of this invention ensures its long-term application in complex and ever-changing industrial environments. As production conditions change, the system can continuously learn and optimize, maintaining a high level of performance. This ability to continuously improve is not found in traditional methods and is a key advantage in the context of Industry 4.0 and smart manufacturing.

[0238] It should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A smart production scheduling and early warning method for hardware processing, characterized in that, Includes the following steps: Acquire real-time production scheduling data from the hardware processing production line; preprocess the real-time production scheduling data to obtain preprocessed data; Based on the preprocessed data, dynamic time window division is performed; multi-scale feature extraction is performed on the divided time window data; Quantum-inspired deep autoencoder networks are used to identify anomalous patterns from extracted multi-scale features. Based on the abnormal pattern recognition results, the cause of the abnormality is diagnosed; Generate and push early warning information; And update the model online based on feedback information; The quantum-inspired deep autoencoder network includes: Quantum state attention mechanism, where attention weight w i The calculation formula is: w i =|a i | 2 =|<ψ|U i |ψ>| 2 Where |ψ> represents the quantum state, |i> represents the computational ground state, and α i For complex amplitude, U i For quantum gate operations; fractal convolution module, where the fractal convolution kernel F k The formula for generating it is: F k =IFS(W,k,s) Where IFS is the iterative function system, W is the set of transformation matrices, k is the number of iterations, and s is the initial shape. Topology adaptive connectivity structure, where network connectivity strength C ij The calculation formula is: C ij =σ(W·Φ(X i ,X j )+b) Where Φ is the continuous cohomology function, σ is the activation function, and W and b are learnable parameters; The non-equilibrium quantum annealing optimization module, where parameter updates follow the non-equilibrium quantum dynamics equations: Where ρ is the density matrix, H is the Hamiltonian, and L is the dissipative superoperator; The abnormal pattern recognition step further includes: The abnormality score AS is calculated using the following formula: AS=log(RE)+λ·D W (X,X ref ) Where RE is the reconstruction error, and D W Let X be the Wasserstein distance. ref As a reference to normal samples, λ is a balance factor; Based on adaptive threshold T adaptive Anomalies are identified, where T adaptive The update formula is: T adaptive =β·T adaptive +(1-β)·(μ(AS)+γ·σ(AS)) Where μ and σ are the mean and standard deviation functions, respectively, β is the smoothing factor, and γ is the sensitivity parameter.

2. The method according to claim 1, characterized in that, The preprocessing steps include: An improved Hampel filter is used to remove outliers, where the filter is calculated using the following formula: Where, x′ i For the filtered data points, x i For the original data points, X w For x i The sliding window is centered, MAD is the absolute deviation of the median, and γ is an adjustable parameter.

3. The method according to claim 1, characterized in that, The dynamic time window division steps include: The formula for calculating the adaptive window size W is as follows: Where ACF is the autocorrelation function and threshold is the preset threshold; an overlapping sliding window sequence is generated based on the adaptive window size W and the preset overlap rate r.

4. The method according to claim 1, characterized in that, The multi-scale feature extraction step includes: The formula for calculating the sample entropy SE(τ) at multiple time scales τ is as follows: Where A m (r) represents the probability of template matching in the m-dimensional phase space under tolerance r; Multi-scale frequency domain features (WPT) obtained by applying wavelet packet transform i The calculation formula is as follows: WPT i =WPT(S i ,level=3,wavelet=′db4′) Where S i The input signal is denoted by 'level', the decomposition level is denoted by 'level', and the wavelet is denoted by 'wavelet', representing the wavelet basis functions. Morphological features MG(S) are extracted using the morphological gradient operator. i The calculation formula is as follows: in and These represent expansion and corrosion operations, respectively, with B being the structural element.

5. The method according to claim 1, characterized in that, The abnormality diagnosis steps include: Feature importance φ is calculated using the Shapley value-based SHAP method. i The calculation formula is as follows: Where F is the feature set, and v is the value function of the feature subset; A causal graph G(V,E) is constructed based on the PC algorithm, where V is the set of nodes and E is the set of edges. A dynamic Bayesian network is used to simulate the propagation of anomalies in the causal graph, and its conditional probability distribution is as follows: P(X t |Pa(X t ))=f(X t ,Pa(X t ),θ) Where Pa(X) t ) is X t The set of parent nodes, f is the conditional probability distribution function, and θ is the model parameter; calculate the parent node X for each node. i Root cause score RCS(X) i The calculation formula is as follows: Where P(path|G,AS) is the probability that an anomaly propagates through path path given an anomaly score AS.

6. The method according to claim 1, characterized in that, The steps of generating and pushing early warning information include: The Comprehensive Severity Index (SDI) is calculated using the following formula: SDI=w1·normalize(AS)+w2·max(normalize(RCS)) Where w1 and w2 are weighting factors, normalize is the normalization function, AS is the anomaly score, and RCS is the root cause score. Generate JSON-formatted alert information that includes timestamps, anomaly scores, severity indices, affected components, root causes and their importance and causal probability, and recommended actions; Early warning information is pushed based on a multi-level push strategy, including: When SDI <T low At that time, only log entries are recorded; When T low ≤SDI <T high At that time, it is pushed to the monitoring system; When SDI≥T high In such cases, information is immediately pushed to relevant personnel through multiple channels.

7. The method according to claim 1, characterized in that, The online model update steps include: Incremental learning is performed using an elastic weight merging algorithm based on the Fisher information matrix, with the loss function L(θ) as follows: Where L new For the loss function of the new data, F i θ represents the diagonal elements of the Fisher information matrix. * The ADWIN algorithm is used to detect concept drift for the old model parameters. For any partition (W0, W1) of the sliding window W, if the following conditions are met: |μ(W0)-μ(W1)|>∈ cut Concept drift was detected, in which m is the window size, and δ is the confidence level parameter; Based on the concept drift detection results, neural architecture search technology is used to dynamically adjust the model structure, with the following optimization objective: Where a represents the architecture, w represents the weights, R represents the reward function, and π represents the architecture distribution strategy. An improved knowledge distillation technique is applied to transfer knowledge from the old model to the new model, with the loss function L... KD for: L Kd =α·L CE (y,σ(z s / T))+(1-a)·L KL (σ(z t / T),σ(z s / T)) Where L CE For cross-entropy loss, L KL Let z be the KL divergence. t and z s α represents the logits for the teacher and student models, respectively, where T is the temperature parameter and α is the equilibrium factor.

8. A hardware processing intelligent scheduling and early warning system for implementing the method of any one of claims 1-7, characterized in that, include: The data acquisition module is used to acquire real-time production scheduling data on the hardware processing production line; The preprocessing module is used to preprocess the real-time production scheduling data; The time window segmentation module is used to perform the dynamic time window segmentation; The feature extraction module is used to perform the multi-scale feature extraction. Anomaly detection module, used to perform anomaly pattern detection using the quantum-inspired deep autoencoder network; The diagnostic module is used to diagnose the cause of the abnormality. The early warning module is used to generate and push the early warning information; as well as The update module is used to update the model online.

Images