A Dynamic Learning Rate Prediction Method Based on EEG Signal and Motion Analysis
By collecting and analyzing EEG signals and work videos, a dynamic learning rate prediction model was constructed, which solved the problem of large prediction deviations in traditional learning curve models in flexible production, and achieved accurate work hour prediction and production scheduling optimization.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- DALIAN JIAOTONG UNIVERSITY
- Filing Date
- 2026-01-08
- Publication Date
- 2026-06-02
AI Technical Summary
Traditional learning curve models assume that the learning rate is a static value, which cannot reflect the nonlinear changes in cognitive state caused by the complexity and variability of tasks, operator fatigue, or fluctuations in attention. Furthermore, they lack quantitative feedback on internal physiological states such as mental load, resulting in large prediction biases and poor adaptability in flexible production.
By simultaneously collecting the operator's EEG signals and operation videos, the complexity of the movements, the frequency of attention, and the time factor are quantitatively analyzed. Based on orthogonal experimental design and multiple linear regression, a dynamic learning rate mapping function is constructed to form the EEG-MOD dynamic learning rate prediction model, which enables real-time response to the operator's physiological cognitive state.
It enables accurate and adaptive prediction of work hours, improves the reliability of production efficiency assessment and production scheduling, and significantly reduces the management challenges caused by learning rate fluctuations.
Smart Images
Figure CN122132821A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of interdisciplinary technology of intelligent manufacturing and human factors engineering, and in particular to a dynamic learning rate prediction method based on electroencephalogram (EEG) signals and motion analysis. Background Technology
[0002] In manufacturing, learning curve models (such as the Wright model and the DeJong model) are core tools for predicting work hours and evaluating production efficiency. Traditional models are mainly based on the static historical data relationship of "cumulative output - work hours per unit" and use modular parameters for empirical fitting and application to predict future work times.
[0003] Because traditional learning curve models assume the learning rate to be a static value, they cannot reflect the nonlinear changes in cognitive state caused by the complexity and variability of tasks, operator fatigue, or fluctuations in attention. At the same time, the models rely entirely on external working hour data and lack quantitative feedback on internal physiological states such as mental load. As a result, when faced with modern flexible production with frequent product changes and varied action sequences, they have large prediction biases and poor adaptability. Summary of the Invention
[0004] To address the problems mentioned in the background art, this invention proposes a dynamic learning rate prediction method based on EEG signal and action analysis. By simultaneously collecting the operator's EEG signals and work videos, the complexity of actions, frequency of attention, and time factors are quantitatively analyzed. Based on orthogonal experimental design and multiple linear regression, a dynamic learning rate mapping function and prediction model that can respond in real time to changes in the operator's physiological and cognitive state are constructed, thereby achieving high-precision adaptive prediction of work hours and improving the reliability of production efficiency assessment and production scheduling.
[0005] To achieve the above-mentioned objectives, the present invention employs the following technical solution:
[0006] In some embodiments of this application, a dynamic learning rate prediction method based on EEG signal and action analysis is provided, comprising the following steps:
[0007] S1. Synchronous acquisition and preprocessing of multimodal data: When the operator is performing a task, the operator's EEG signal and video data of the task process are acquired simultaneously. The EEG signal is processed for artifact removal, filtering and task segmentation. The video data is processed for action segmentation and keyframe marking.
[0008] S2. Quantitative analysis of key influencing factors: Based on the collected EEG signals and video data, the complexity of the action ε, the frequency coefficient of attention φ, and the operation time T are quantitatively calculated.
[0009] S3. Key Factor Influence Analysis Based on Orthogonal Experiment: Design an orthogonal experiment, using the energy characteristics of the EEG signal as the response variable, to analyze the significance of the influence of action complexity ε, attention frequency φ, and operation time T on the learning rate;
[0010] S4. Construction of the dynamic learning rate mapping model: Based on the results of orthogonal experiments, a multiple regression model is established with action complexity ε, attention frequency coefficient φ, and operation time T as independent variables and learning rate as dependent variable, to obtain the dynamic learning rate function f(ε,φ,T).
[0011] S5. Integration and application of dynamic learning rate prediction model: The dynamic learning rate function f(ε,φ,T) is integrated into the learning curve model to form the EEG-MOD dynamic learning rate prediction model. The parameters are dynamically calculated and subsequent working hours are predicted based on the real-time collected EEG signals and video data of the work process.
[0012] In some embodiments of this application, in step S1, a portable or multi-channel EEG device is used to acquire the EEG signal, and a wide-angle high-definition camera is used to acquire the video data of the operation process.
[0013] In some embodiments of this application, in step S2, the quantitative calculation formula for the action complexity ε is:
[0014]
[0015] in, The standard time coefficient of element i. The number of times element i appears in the action sequence is given, and the element is obtained by parsing based on a predetermined action time standard method.
[0016] In some embodiments of this application, in step S2, the attention frequency coefficient φ is calculated based on a binomial distribution model, and the calculation formula is as follows:
[0017]
[0018] Where n is the total number of action sequences, k is the number of times an attentional motif appears, and P is the probability of an attentional motif appearing.
[0019] In some embodiments of this application, step S2 further includes standardizing the operation time T. The standardization of the operation time T includes recording the actual time T for completing a standard work unit. 实 And standardized according to the following formula:
[0020]
[0021] Forecast time mean The calculation formula is:
[0022]
[0023] In some embodiments of this application, in step S3, the energy characteristics of the EEG signal are obtained by performing a Fourier transform on the EEG signal, calculating the energy density of each frequency band, and synthesizing a comprehensive component. The energy density calculation formula is as follows:
[0024]
[0025] In some embodiments of this application, in step S4, the dynamic learning rate function is:
[0026]
[0027] In some embodiments of this application, in step S5, the EEG-MOD dynamic learning rate prediction model is an improved DeJong learning curve model, with the following expression:
[0028]
[0029] Among them, C x Let C1 be the predicted working time for the Xth product, C1 be the working time for the first product, and M be the machine intervention coefficient.
[0030] In some embodiments of this application, the dynamic learning rate prediction method is applicable to repetitive operation tasks in a multi-variety, small-batch production environment.
[0031] In some embodiments of this application, the dynamic learning rate prediction method further includes integrating the EEG-MOD dynamic learning rate prediction model into a manufacturing execution system or a digital twin platform for real-time production scheduling and job optimization.
[0032] Compared with the prior art, the advantages and positive effects of the present invention are:
[0033] This invention collects and analyzes the operator's EEG signals in real time, and combines this with quantitative analysis of action complexity ε, attention frequency φ, and operation time T to construct a dynamic learning rate model that can dynamically respond to changes in the operator's physiological and task states. This enables accurate and adaptive prediction of work hours and significantly improves the reliability of production efficiency assessment and production scheduling.
[0034] This invention realizes the perception and dynamic modeling of the operator's physiological state during the learning process, transforming the traditional static and passive time prediction into a dynamic, adaptive, and accurate prediction based on physiological feedback, effectively solving the management problems caused by learning rate fluctuations in flexible production.
[0035] Other features and advantages of the present invention will become clearer after reading the detailed embodiments of the invention in conjunction with the accompanying drawings. Attached Figure Description
[0036] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0037] Figure 1 A flowchart for constructing the EEG-MOD learning curve according to some embodiments;
[0038] Figure 2 A graph showing the fitting of principal components of brainwaves to the learning rate according to some embodiments;
[0039] Figure 3 EEG-MOD learning curve according to some embodiments;
[0040] Figure 4 This is a flowchart illustrating the wiring process for an instrument box according to some embodiments;
[0041] Figure 5 Box plots showing learning rates compared to some embodiments. Detailed Implementation
[0042] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0043] The following disclosure provides many different embodiments or examples for implementing various structures of the invention. To simplify the disclosure, specific examples of components and arrangements are described below. These are merely examples and are not intended to limit the invention. Furthermore, reference numerals and / or letters may be repeated in different examples; such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed. In addition, examples of various specific processes and materials are provided in this invention, but those skilled in the art will recognize the application of other processes and / or the use of other materials.
[0044] The learning curve refers to the relationship between the time an operator spends producing a single unit and the production volume during mass production. With increasingly diverse and personalized customer demands, the multi-variety, small-batch production model places higher demands on operators' dynamic learning abilities.
[0045] The model proposed by Garg and Milliman is an empirical formula primarily used to describe the relationship between learning rate and learning amount during the learning process. The model takes the following form:
[0046]
[0047] Where: X represents the learning rate; n represents the learning amount, that is, the amount of information already learned during the learning process; K, a, and b are constants used to adjust the shape and scale of the model.
[0048] The DeJong J model incorporates a human-machine coefficient M into the traditional learning curve. The mathematical expression for this model is:
[0049]
[0050] When M=1, it means that the processing is entirely done by machines, without human intervention. When human-machine collaboration is involved, the learning rate of the proficiency curve is a variable value.
[0051] Because traditional learning curve models assume the learning rate to be a static value, they cannot reflect the nonlinear changes in cognitive state caused by the complexity and variability of tasks, operator fatigue, or fluctuations in attention. At the same time, the models rely entirely on external working hour data and lack quantitative feedback on internal physiological states such as mental load. As a result, when faced with modern flexible production with frequent product changes and varied action sequences, they have large prediction biases and poor adaptability.
[0052] Based on this, this solution proposes a dynamic learning rate prediction method based on EEG signal and motion analysis. Its core lies in constructing a quantitative model that can dynamically reflect changes in the operator's cognitive state and learning efficiency by synchronously collecting the operator's EEG signals and motion information.
[0053] Specifically, in some embodiments of this application, a method for predicting dynamic learning rates based on EEG signals and action analysis is provided, including the following steps:
[0054] S1. Synchronous acquisition and preprocessing of multimodal data;
[0055] S2. Quantitative analysis of key influencing factors;
[0056] S3. Analysis of the Influence of Key Factors Based on Orthogonal Experiments;
[0057] S4. Construction of the dynamic learning rate mapping model;
[0058] Integration and application of S5 and EEG-MOD dynamic learning rate prediction models.
[0059] Among them, reference Figure 1:
[0060] Step S1, Multimodal Data Synchronous Acquisition and Preprocessing, includes: when the operator is performing a task, such as assembly or wiring, the operator's EEG signal and video data of the task process are acquired simultaneously; the EEG signal is processed for artifact removal, filtering (e.g., 0.5-45Hz) and task segmentation; and the video data is processed for action segmentation and keyframe marking to provide input for subsequent action analysis.
[0061] Step S2, Quantitative Analysis of Key Influencing Factors, includes: Quantitatively calculating the action complexity ε, attention frequency coefficient φ, and operation time T based on the collected EEG signals and video data. Action complexity ε, attention frequency coefficient φ, and operation time T are three factors affecting the learning rate.
[0062] Step S3, the key factor influence analysis based on orthogonal experiments, includes: To scientifically determine the significance of the above three factors' influence on the learning rate, an orthogonal experiment is designed and executed. Using the energy characteristics of the EEG signals as the response variable, the significance of the influence of action complexity ε, attention frequency φ, and operation time T on the learning rate is analyzed.
[0063] Step S4, the construction of the dynamic learning rate mapping model, includes: based on the results of orthogonal experiments, establishing a multiple regression model with action complexity ε, attention frequency coefficient φ, and operation time T as independent variables and learning rate as dependent variable, to obtain the dynamic learning rate function f(ε,φ,T).
[0064] Step S5, the integration and application of the dynamic learning rate prediction model, includes: integrating the dynamic learning rate function f(ε,φ,T) into the learning curve model to form the EEG-MOD dynamic learning rate prediction model, and dynamically calculating parameters and predicting subsequent working hours based on the real-time collected EEG signals and video data of the work process.
[0065] This invention collects and analyzes the operator's EEG signals in real time, and combines this with quantitative analysis of action complexity ε, attention frequency φ, and operation time T to construct a dynamic learning rate model that can dynamically respond to changes in the operator's physiological and task states. This enables accurate and adaptive prediction of work hours and significantly improves the reliability of production efficiency assessment and production scheduling.
[0066] This invention realizes the perception and dynamic modeling of the operator's physiological state during the learning process, transforming the traditional static and passive time prediction into a dynamic, adaptive, and accurate prediction based on physiological feedback, effectively solving the management problems caused by learning rate fluctuations in flexible production.
[0067] In some embodiments of this application, in step S1, a portable or multi-channel EEG device is used to acquire the EEG signals, and electrodes are placed according to a standard lead layout (e.g., the international 10-20 system) to acquire the operator's raw EEG signals in real time during the task.
[0068] The operation process was captured using a wide-angle high-definition camera. The camera was positioned above the work station to ensure it could record the operator's hands working, guaranteeing clear images and coverage of the entire range of motion.
[0069] In some embodiments of this application, in step S2, the work actions are analyzed based on the MOD method, i.e., the predetermined action time standard method, and the action sequence is decomposed into standard motion elements, such as M1, M2, ..., M5, etc. The quantitative calculation formula for the action complexity ε is:
[0070]
[0071] in, The standard time coefficient of element i, for example, M1=1 MOD. The number of times motion element i appears in the motion element sequence is given by the motion element, which is obtained by parsing based on a predetermined action time standard method.
[0072] In some embodiments of this application, in step S2, attentional elements that require visual attention or decision-making in the task are identified, such as seeing the direction E2D3.
[0073] The attention frequency coefficient φ is calculated based on a binomial distribution model, and the calculation formula is as follows:
[0074]
[0075] Where n is the total number of action sequences, k is the number of times an attentional motif appears, and P is the probability of an attentional motif appearing.
[0076] In some embodiments of this application, step S2 further includes standardizing the operation time T to eliminate the influence of individual speed differences. The standardization of the operation time T includes recording the actual time T for completing a standard work unit (e.g., a process cycle). 实 And standardized according to the following formula:
[0077]
[0078] The predicted time mean is shown in the following formula:
[0079]
[0080] In some embodiments of this application, in step S3, the energy characteristics of the EEG signal are obtained by performing a Fourier transform on the EEG signal, calculating the energy density of each frequency band, and synthesizing a comprehensive component. The energy density calculation formula is as follows:
[0081]
[0082] In some embodiments of this application, to facilitate the analysis and better representation of EEG energy, the energy is converted to a logarithmic scale for simplification:
[0083]
[0084] In some embodiments of this application, three factors were set: action complexity ε, attention frequency φ, and operation time T, with three levels for each factor tested. Range analysis and variance analysis were used to determine the order of significance of each factor's influence on the learning rate. The results showed that action complexity ε had the most significant impact on EEG changes (F=4.77, p=0.030), and was the core factor driving the dynamic changes in the learning rate.
[0085] The real-time learning rate S for each task segment is calculated using the Wright learning curve formula:
[0086]
[0087]
[0088] The fitting results of the EEG component data corresponding to the action and the learning rate data are as follows: Figure 2 As shown.
[0089] In some embodiments of this application, in step S4, a quantitative mapping relationship between EEG features and learning rate is established based on the results of orthogonal experiments.
[0090] First, based on Garg and Milliman's empirical formula and combined with MOD theory, empirical model assumptions are made:
[0091]
[0092] Then, using action complexity ε, attention frequency coefficient φ, and operation time T as independent variables and learning rate as the dependent variable, multiple linear regression was used for model fitting. According to the statistical results of the regression model, the goodness of fit R0 is [value missing]. 2 =0.89 indicates good explanatory power.
[0093] The final dynamic learning rate function is:
[0094]
[0095] In some embodiments of this application, in step S5, the aforementioned dynamic learning rate function is... This is integrated into the improved DeJong learning curve model to form the final EEG-MOD dynamic prediction model, such as... Figure 3 As shown.
[0096] The EEG-MOD dynamic learning rate prediction model is an improved DeJong learning curve model, and its expression is as follows:
[0097]
[0098] Among them, C x Let C1 be the predicted working time for the Xth product, C1 be the working time for the first product, and M be the machine intervention coefficient.
[0099] In practical applications, the system can dynamically calculate current parameters by collecting the operator's EEG and movement data in real time and inputting them into the above model, thereby enabling adaptive prediction of subsequent working hours and providing real-time decision support for production scheduling.
[0100] In some embodiments of this application, the dynamic learning rate prediction method is applicable to repetitive operation tasks in a multi-variety, small-batch production environment.
[0101] In some embodiments of this application, the dynamic learning rate prediction method further includes integrating the EEG-MOD dynamic learning rate prediction model into a manufacturing execution system or a digital twin platform for real-time production scheduling and job optimization.
[0102] In some embodiments of this application, the application of the dynamic learning rate prediction method in the wiring process of low-voltage distribution cabinets is given.
[0103] The process flow diagram for wiring network circuit breakers and aviation sockets in a company's low-voltage distribution cabinet is as follows: Figure 4 As shown. Based on real-time action analysis, the action complexity ε, attention frequency coefficient φ, and operation time T at different times are calculated and substituted into the EEG-MOD model to dynamically predict the learning rate.
[0104] Compared to traditional static learning curves, the learning rate bias range predicted by the model of this invention is significantly reduced, and can be controlled within ±3%, demonstrating its excellent effect in improving the accuracy of time prediction in real complex operations. Figure 5 As shown.
[0105] In the description of the above embodiments, specific features, structures, materials, or characteristics may be combined in any suitable manner in one or more embodiments or examples.
[0106] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.
Claims
1. A dynamic learning rate prediction method based on EEG signal and motion analysis, characterized in that, It includes the following steps: S1. Synchronous acquisition and preprocessing of multimodal data: When the operator is performing a task, the operator's EEG signal and video data of the task process are acquired simultaneously. The EEG signal is processed for artifact removal, filtering and task segmentation. The video data is processed for action segmentation and keyframe marking. S2. Quantitative analysis of key influencing factors: Based on the collected EEG signals and video data, the complexity of the action ε, the frequency coefficient of attention φ, and the operation time T are quantitatively calculated. S3. Key Factor Influence Analysis Based on Orthogonal Experiment: Design an orthogonal experiment, using the energy characteristics of the EEG signal as the response variable, to analyze the significance of the influence of action complexity ε, attention frequency φ, and operation time T on the learning rate; S4. Construction of the dynamic learning rate mapping model: Based on the results of orthogonal experiments, a multiple regression model is established with action complexity ε, attention frequency coefficient φ, and operation time T as independent variables and learning rate as dependent variable, to obtain the dynamic learning rate function f(ε,φ,T). S5. Integration and application of dynamic learning rate prediction model: The dynamic learning rate function f(ε,φ,T) is integrated into the learning curve model to form the EEG-MOD dynamic learning rate prediction model. The parameters are dynamically calculated and subsequent working hours are predicted based on the real-time collected EEG signals and video data of the work process.
2. The dynamic learning rate prediction method according to claim 1, characterized in that, In step S1, a portable or multi-channel EEG device is used to acquire the EEG signals, and a wide-angle high-definition camera is used to acquire the video data of the operation process.
3. The dynamic learning rate prediction method according to claim 1, characterized in that, In step S2, the formula for quantifying the action complexity ε is: ； in, The standard time coefficient of element i. The number of times motion element i appears in the action sequence is given, and the motion element is obtained based on a predetermined action time standard analysis method.
4. The dynamic learning rate prediction method according to claim 1, characterized in that, In step S2, the attention frequency coefficient φ is calculated based on a binomial distribution model, and the calculation formula is as follows: ； Where n is the total number of times the action sequence occurs, k is the number of times the attentional motif appears, and p is the probability of the attentional motif appearing.
5. The dynamic learning rate prediction method according to claim 1, characterized in that, Step S2 further includes standardizing the operation time T, which includes recording the actual time T for completing one standard work unit. 实 And standardized according to the following formula: ； Forecast time mean The calculation formula is: 。 6. The dynamic learning rate prediction method according to claim 1, characterized in that, In step S3, the energy characteristics of the EEG signal are obtained by performing a Fourier transform on the EEG signal, calculating the energy density of each frequency band, and synthesizing a comprehensive component. The energy density calculation formula is as follows:
7. The dynamic learning rate prediction method according to claim 1, characterized in that, In step S4, the dynamic learning rate function is: 。 8. The dynamic learning rate prediction method according to claim 1, characterized in that, In step S5, the EEG-MOD dynamic learning rate prediction model is an improved DeJong learning curve model, with the following expression: ； Among them, C x Let C1 be the predicted working time for the Xth product, C1 be the working time for the first product, and M be the machine intervention coefficient.
9. The dynamic learning rate prediction method according to any one of claims 1 to 8, characterized in that, The dynamic learning rate prediction method is applicable to repetitive tasks in multi-variety, small-batch production environments.
10. The dynamic learning rate prediction method according to any one of claims 1 to 8, characterized in that, The dynamic learning rate prediction method also includes integrating the EEG-MOD dynamic learning rate prediction model into a manufacturing execution system or digital twin platform for real-time production scheduling and job optimization.