Method for analyzing indexes of production machine and whole line based on Internet of Things technology
Through the multi-dimensional evaluation model and the space-time convolution coupled graph attention network, combined with the dynamic fatigue attenuation coefficient and intelligent allocation decision model, the problems of workers' ability judgment and machine equipment failure prediction on the production line are solved, and efficient, low-cost management and stable operation of the production line are achieved.
Patent Information
- Application Number
- CN202510553641.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-29
- Publication Date
- 2025-08-05
AI Technical Summary
The existing production lines have problems such as lack of objective evaluation of workers' work ability judgment and poor accuracy of machine equipment failure prediction, which makes it difficult to ensure production efficiency and output.
By building a multi-dimensional evaluation model and a spatiotemporal convolution coupled graph attention network, combining dynamic fatigue attenuation coefficients and intelligent allocation decision model, workers' capacity quantification and machine equipment failure prediction are realized, and intelligent workers' allocation is carried out to optimize production line management.
It realizes accurate quantitative judgment of workers' abilities and accurate failure prediction of machine equipment, improves the efficiency and output of the production line, reduces management costs, and ensures stable operation and yield of the production line.
Smart Images
Figure CN120430683A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of production line management, and in particular to a method for analyzing indicators of production machines and entire lines based on Internet of Things technology. Background Art
[0002] Although existing production lines use automated processing equipment, manual operations such as loading and unloading are still required. Conventionally, one or more workers are often deployed on a production line. For the sake of production efficiency, on-site supervision is often mainly dependent on manual supervision, which will undoubtedly increase labor management costs. Production efficiency is affected by the workers' work ability. Traditional judgments on workers' work ability are highly subjective and often rely on subjective scores of managers. There is a lack of effective quantitative standards, resulting in the inability to guarantee the smooth completion of the production line's target output when the production line increases output demand. At the same time, during the operation of the production line, machine equipment is prone to malfunction, which seriously affects the completion of the target output. Traditional fault prediction of the production line tends to ignore the temporal and spatial correlation, affecting the accuracy of fault prediction.
[0003] Based on the retrieval of the above information, a method for analyzing the indicators of production machines and entire lines based on Internet of Things technology is proposed. The method collects statistics on the output efficiency of production lines, product quality and the response efficiency of production line machines and equipment, and makes multi-dimensional judgments on the comprehensive ability indicators of workers in combination with their work numbers. This method provides accurate data support for the use and deployment of workers required by the production line, and intelligently allocates the types of workers required by the production line. On the basis of ensuring the yield rate and output, low-cost and efficient management of the production line is achieved. Summary of the Invention
[0004] In response to the shortcomings of the existing technology, the present invention provides a method for analyzing indicators of production machines and entire lines based on Internet of Things technology, which solves the problems of lack of objective evaluation of workers' work ability and poor accuracy in predicting machine equipment failures.
[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: a method for analyzing indicators of production machines and entire lines based on Internet of Things technology, specifically comprising the following steps: Step 1: Capacity Positioning: Collect machine output, worker ID, and output quality data, build a multi-dimensional evaluation model, and calculate the worker's comprehensive capacity index; Step 2: Fault prediction: Build a spatiotemporal convolutional coupled graph attention network to predict faults of production line equipment; Step 3: Intelligent Allocation: Build an intelligent allocation decision model to allocate workers to corresponding production lines with the optimization goal of minimizing costs and losses; Step 4: Entropy warning: Calculate the entropy value of the worker ability distribution and issue an abnormal alarm when it is lower than the preset threshold.
[0006] The present invention is further configured as follows: the multi-dimensional evaluation model in step 1 is: Where, For the The comprehensive ability index score of each worker, For the The output per unit time of a worker, is the average output per unit time of the production line, For the The yield rate of each worker, For the Average fault response time per worker, is the response speed attenuation coefficient, and , For the The ability stability coefficient of each worker, are weight coefficients, and .
[0007] The present invention is further configured as follows: the calculation formula of the spatiotemporal convolution coupled graph attention network is: Where, For the Quantile failure rate prediction value, is the weight matrix of the quantile regression model, is the feature fusion function, is the temporal convolution feature, Attention features for spatial graphs, For environmental integration features, is the bias term.
[0008] The present invention is further configured as follows: when performing fault prediction for all equipment on the production line in step 2, the fatigue attenuation coefficient is dynamically updated according to the accumulated working hours, and the formula is: Where, is the fatigue attenuation coefficient, Accumulate working hours for machine equipment, is the mean time between failures, is the fatigue attenuation factor, is a nonlinear index.
[0009] The present invention is further configured such that the fatigue attenuation coefficient is used to update the failure rate prediction value, including: Where, is the corrected failure rate prediction value, is the original failure rate prediction value, is the baseline failure rate, It is the theoretical mean time between failures of the equipment.
[0010] The present invention is further configured as follows: the intelligent deployment decision model includes: Where, For the Is the worker assigned to A binary variable for each production line, For the The equipment maintenance intensity of each production line, For the The labor cost per worker per unit time, For the The unit time equipment maintenance cost of a production line, is the mass loss base, To control quality sensitivity, To set the production target, is the dynamic demand coefficient (usually 0.8-1.2), is the minimum yield constraint, Capacity for workers and production line requirements The Pearson correlation coefficient, is the minimum correlation threshold.
[0011] The present invention is further configured such that: the intelligent deployment decision model is solved by a quantum heuristic algorithm, and the Pareto frontier solution search is accelerated by quantum chromosome encoding, including: Where, is the quantum chromosome state vector, N is the number of quantum bits, that is, the number of workers, For the The rotation angle of the qubits (coding worker-production line allocation scheme), and is the ground state of the quantum bit.
[0012] The present invention is further configured as follows: the formula for calculating the entropy value of worker ability distribution is: Where, is the entropy of worker ability distribution, For the The proportion of workers with different ability levels, Assign quantities to the capability levels.
[0013] The present invention provides a method for analyzing indicators of production machines and entire production lines based on Internet of Things technology. It has the following beneficial effects: The present invention combines actual production data to conduct a multi-dimensional comprehensive analysis of workers' work capabilities, realizes quantitative judgment of workers' comprehensive capabilities, cooperates with the spatiotemporal convolution coupling of production line equipment to perform fault prediction, and realizes accurate fault prediction of equipment through dynamic updating of dynamic fatigue attenuation coefficient, providing reliable guarantee for the completion of production line target output. Then, workers are intelligently deployed according to their comprehensive capabilities and production line target output, combined with yield rate constraints, to realize low-cost and efficient output management of production lines, and through the calculation of entropy value of workers' ability distribution, provide reliable guarantee for stable operation of production lines. BRIEF DESCRIPTION OF THE DRAWINGS
[0014] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION
[0015] The technical solutions in the embodiments of the present invention will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present invention.
[0016] See also Figure 1 The embodiment of the present invention provides the following technical solution: a method for analyzing indicators of production machines and entire lines based on Internet of Things technology, specifically comprising the following steps: Step 1: Capacity Positioning: Collect machine output, worker ID, and output quality data, build a multi-dimensional evaluation model, and calculate the worker's comprehensive capability indicators, including: Where, For the The comprehensive ability index score of each worker, For the The output per unit time of a worker, is the average output per unit time of the production line, For the The yield rate of each worker, is the average yield rate of the production line, For the Average fault response time per worker, is the response speed attenuation coefficient, and , For the The ability stability coefficient of each worker, are weight coefficients, and ,Regarding the weight coefficient, the hierarchical analysis method is used to ,determine the initial weight, and then the weight coefficient is updated quarterly or ,annually based on the following entropy value of workers’ ability ,distribution.
[0017] The average output per unit time of the production line and the average yield rate of the production line are updated once according to a preset time period, including but not limited to daily and weekly.
[0018] To further explain, for the output per unit time, sensors are used to collect the output quantity of the equipment in real time. The sensors include but are not limited to photoelectric sensors and weighing sensors. The total output during the operating time is divided by the working time to obtain the output per unit time.
[0019] To further explain, the average fault response time is calculated based on the fault occurrence time and response start time recorded in the equipment fault log, and the time when the worker arrives at the equipment by scanning the wireless radio frequency tag or work badge. The formula is: Where, is the response start time, The time when the fault occurred.
[0020] To further illustrate, for the worker's ability stability coefficient, we collect at least 30 days of historical output data of the worker and use a sliding window to calculate the output volatility. The formula is: Where, is the standard deviation of historical output data, is the mean of historical output data, and stability is graded as A, B, and C, where: A-level: , indicating that the ability is highly stable; B-level: , indicating stable ability C-level: , indicating that the ability fluctuates greatly.
[0021] Step 2: Fault prediction: Build a spatiotemporal convolution coupled graph attention network. The spatiotemporal convolution coupled graph attention network architecture includes: Temporal convolution layer: Three layers of dilated convolution with dilation rates of 1, 2, and 4, respectively. Dilated convolution is used to capture the temporal dependency of device states, and multi-scale dilation rates are used to obtain features at different temporal granularities. Spatial graph attention layer: Models production line equipment as a graph structure, with nodes representing equipment and edges representing physical connections or functional associations between equipment. A four-head attention mechanism is used to learn the fault propagation patterns of equipment. Environmental fusion layer: The fully connected layer processes the machine equipment operating status data and performs feature fusion. The machine equipment operating status data includes but is not limited to temperature, humidity, and vibration data.
[0022] The spatiotemporal convolution coupled graph attention network is used to predict the faults of production line equipment. The calculation formula is: Where, For the Quantile failure rate prediction value, is the weight matrix of the quantile regression model, is the feature fusion function, is the temporal convolution feature, Attention features for spatial graphs, For environmental integration features, is the bias term.
[0023] Further explanation: In order to improve the accuracy of fault prediction, the fatigue attenuation coefficient is dynamically updated according to the accumulated working hours. The formula is: Where, is the fatigue attenuation coefficient, updated in hours. Accumulate working hours for machine equipment, is the mean time between failures, is the fatigue attenuation factor, determined by Bayesian optimization, and taken as 0.05-0.15. It is a nonlinear exponent that controls the shape of the attenuation curve and is set between 1.2 and 1.8. The fatigue attenuation coefficient is used to update the failure rate prediction value, including: Where, is the corrected failure rate prediction value, is the original failure rate prediction value, is the baseline failure rate, It is the theoretical mean time between failures of the equipment.
[0024] During use, the cumulative working hours of the equipment are updated every hour, and the 72-hour sliding window data is used to train the spatiotemporal convolution coupled graph attention network. The specific implementation of the spatiotemporal convolution coupled graph attention network includes: A1. Data Collection: IoT sensors collect real-time data on the operating status of equipment, and obtain historical fault records and maintenance logs of equipment; A2. Feature Engineering: Construct time series features, generate equipment association graphs, and standardize equipment operation status data; A3. Model training: The historical data was divided into a training set, a validation set, and a test set in a ratio of 14:3:3. The Adam optimizer was used with a learning rate of 0.001. The multi-task learning framework was used to jointly optimize the quantile regression objective. A4. Online prediction: Read the accumulated working hours of the machine equipment in real time, calculate the dynamic fatigue attenuation coefficient, and output the corrected failure rate prediction value.
[0025] Step 3: Intelligent Allocation: Build an intelligent allocation decision model to allocate workers to the production line with the optimization goal of minimizing costs and losses. The objective function of the intelligent allocation decision model is: Where, For the Is the worker assigned to A binary variable for each production line, For the The equipment maintenance intensity of each production line is a normalized value in the range of 0-1. For the The labor cost per worker per unit time, For the The unit time equipment maintenance cost of a production line, is the mass loss base, with a value of 0.1-0.5, is the mass sensitivity, with a value of 0.05-2, The comprehensive yield rate of products produced by all machines and equipment in the current production line; The intelligent allocation decision model adopts production constraints, quality constraints and capacity matching constraints, among which the production constraints are: Where, To set the production target, is the dynamic demand coefficient, with a value of 0.8-1.2; The quality constraints are: is the minimum yield constraint; The capability matching constraints are: Capacity for workers and production line requirements The Pearson correlation coefficient, is the minimum correlation threshold.
[0026] As a preferred solution, the intelligent allocation decision model is solved through a quantum heuristic algorithm, encoded through quantum chromosomes, and adopts a quantum rotating gate update strategy to accelerate the search for Pareto frontier solutions, including: Where, is the quantum chromosome state vector, N is the number of quantum bits, that is, the number of workers, For the The rotation angle of the quantum bit encodes the worker-production line allocation relationship, and is the ground state of the quantum bit.
[0027] The solution process includes: Initialize the population: randomly generate quantum bit rotation angles ; Iterative optimization: Calculate the objective function value of the current solution, update the collar bit state according to the Pareto dominance relationship, adjust the rotation angle, with a step size of 0.1π and a maximum number of iterations of 200. When the maximum number of iterations or the convergence threshold is reached, output the Pareto optimal solution set. The optimal solution set contains multiple non-dominated solutions and provides different cost-quality trade-offs. The final solution can be selected based on decision preferences.
[0028] The method of using the intelligent allocation decision model includes: B1. Data Collection: Real-time reading of worker status (on duty / resting) and production line equipment status (operating / maintenance), obtaining the latest order demand and dynamic demand coefficient; B2. Model input: Input the list of currently available workers into the worker pool and the list of all running production lines into the production line pool to determine the quality loss base, quality sensitivity, and minimum yield constraint; B3. Online solution: Perform global deployment once per shift, using an 8-hour shift. Use a quantum-inspired algorithm to solve the multi-objective optimization model and output the optimal deployment plan and maintenance intensity. B4. Execution and Feedback: Send the deployment plan to the production line management system, monitor actual output data, update model parameters, and recalibrate model parameters every quarter.
[0029] Step 4: Entropy warning: Calculate the entropy of worker ability distribution using the following formula: Where, is the entropy of worker ability distribution, For the The proportion of workers with different ability levels, For the number of ability levels, workers are divided into K ability levels through clustering algorithm, preferably five levels; When the entropy value of the worker ability distribution is lower than the preset threshold, an abnormal alarm is issued, indicating that a skill gap has occurred, and a low-level worker distribution report is generated, prompting worker training plans or worker secondment.
[0030] Simulation experiment To verify the accuracy of quantifying worker capabilities, this paper compares the present invention with traditional subjective evaluation methods and piece-rate methods. The subjective evaluation method uses a scoring evaluation by the team leader, while the piece-rate method uses statistical output. The correlation coefficient of the actual efficiency of the production line is used as the evaluation accuracy indicator, and the proportion of workers with low actual efficiency is used as the low-scoring worker identification rate indicator. The comparison results are shown in Table 1: Evaluation accuracy Low-score worker recognition rate The present invention 0.82 85% Subjective evaluation method 0.42 58% Piecework 0.61 65% Table 1 As can be seen from Table 1, the present invention improves the evaluation accuracy by 41% and the recognition rate of low-score workers by 20% through multi-dimensional quantification.
[0031] To verify the fault prediction accuracy of the present invention, an experimental scenario was conducted on a certain automotive parts stamping production line with 10 devices and 5 years of historical fault records. The LSTM model and the random forest model were used as comparison models, with the root mean square error as the prediction error indicator and the remaining life accuracy covering the 95% confidence interval as the indicator. The comparison results are shown in Table 2: Prediction error Remaining life accuracy The present invention 22.1 92% LSTM model 38.2 68% Random Forest 32.5 75% Table 2 As can be seen from Table 2, the present invention reduces the prediction error by 42% and improves the accuracy of remaining life by 17% through spatiotemporal coupling modeling.
[0032] To verify the optimization effect of the present invention on production line allocation, a semiconductor packaging production line with 50 workers, 20 pieces of equipment, and a dynamic order fluctuation of ±30% was tested using a genetic algorithm and a particle swarm optimization algorithm for comparison. The results of the test were shown in Table 3: Production line efficiency / % Cost reduction rate / % Improvement in yield rate / % The present invention 88 18 2.3 Genetic Algorithm 81 12 1.1 Particle Swarm Optimization 83 15 1.5 Table 3 As can be seen from Table 3, the production line efficiency increased by 7%, the cost reduction rate increased by 3%, and the yield rate increased by 0.8%.
[0033] In summary, the present invention achieves the judgment accuracy of worker capabilities through multi-dimensional quantification, quota and spatiotemporal coupling modeling, effectively reducing fault prediction errors. Combined with intelligent allocation, it achieves the effects of reducing costs while improving yield and production line efficiency. In practical applications, it can improve the overall efficiency of the production line by 20%-30%, and has good application prospects.
[0034] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.
Claims
1. A method for analyzing indicators of production machines and entire lines based on Internet of Things technology, characterized in that: The specific steps include: Step 1: Capacity Positioning: Collect machine output, worker ID, and output quality data, build a multi-dimensional evaluation model, and calculate the worker's comprehensive capacity index; Step 2: Fault prediction: Build a spatiotemporal convolutional coupled graph attention network to predict faults of production line equipment; Step 3: Intelligent Allocation: Build an intelligent allocation decision model to allocate workers to corresponding production lines with the optimization goal of minimizing costs and losses; Step 4: Entropy warning: Calculate the entropy value of the worker ability distribution and issue an abnormal alarm when it is lower than the preset threshold.
2. The method for analyzing indicators of production machines and entire lines based on Internet of Things technology according to claim 1, characterized in that: The multi-dimensional evaluation model in step 1 is: Where, For the The comprehensive ability index score of each worker, For the The output per unit time of a worker, is the average output per unit time of the production line, For the The yield rate of each worker, For the Average fault response time per worker, is the response speed attenuation coefficient, and , For the The ability stability coefficient of each worker, are weight coefficients, and .
3. The method for analyzing indicators of production machines and entire lines based on Internet of Things technology according to claim 1, characterized in that: The calculation formula of the spatiotemporal convolution coupled graph attention network is: Where, For the Quantile failure rate prediction value, is the weight matrix of the quantile regression model, is the feature fusion function, is the temporal convolution feature, Attention features for spatial graphs, For environmental integration features, is the bias term.
4. The method for analyzing indicators of production machines and entire lines based on Internet of Things technology according to claim 3, characterized in that: When performing the fault prediction for all equipment on the production line in step 2, the fatigue attenuation coefficient is dynamically updated according to the accumulated working hours. The formula is: Where, is the fatigue attenuation coefficient, Accumulate working hours for machine equipment, is the mean time between failures, is the fatigue attenuation factor, is a nonlinear index.
5. The method for analyzing indicators of production machines and entire lines based on Internet of Things technology according to claim 3, characterized in that: The fatigue attenuation coefficient is used to update the failure rate prediction value, including: Where, is the corrected failure rate prediction value, is the original failure rate prediction value, is the baseline failure rate, It is the theoretical mean time between failures of the equipment.
6. The method for analyzing indicators of production machines and entire lines based on Internet of Things technology according to claim 1, characterized in that: The intelligent deployment decision model includes: Where, For the Is the worker assigned to A binary variable for each production line, For the The equipment maintenance intensity of each production line, For the The labor cost per worker per unit time, For the The unit time equipment maintenance cost of a production line, is the mass loss base, To control quality sensitivity, To set the production target, is the dynamic demand coefficient (usually 0.8-1.2), is the minimum yield constraint, Capacity for workers and production line requirements The Pearson correlation coefficient, is the minimum correlation threshold.
7. The method for analyzing indicators of production machines and entire lines based on Internet of Things technology according to claim 6, characterized in that: The intelligent deployment decision model is solved by a quantum heuristic algorithm and accelerates the Pareto frontier solution search through quantum chromosome encoding, including: Where, is the quantum chromosome state vector, N is the number of quantum bits, that is, the number of workers, For the The rotation angle of the quantum bit, and is the ground state of the quantum bit.
8. The method for analyzing indicators of production machines and entire lines based on Internet of Things technology according to claim 1, characterized in that: The formula for calculating the entropy value of worker ability distribution is: Where, is the entropy of worker ability distribution, For the The proportion of workers with different ability levels, Assign quantities to the capability levels.