A big data-based efficiency maximization management method for hardware processing workshops

Abstract

This invention discloses a big data-based method for maximizing efficiency management in a hardware processing workshop, relating to the field of big data management technology. The method includes: acquiring historical video data of hardware workpiece processing tasks; quantifying the skill levels of processing personnel in the hardware processing workshop; clustering and classifying the processing personnel according to their skill levels; initializing the time distribution of hardware workpiece processing actions under each skill level; constructing a hardware workpiece processing action-effective / ineffective time identification model; marking the types of ineffective time-consuming tasks for each worker's skill level; performing causal tracing for these ineffective time-consuming task types; constructing a causal list of ineffective time-consuming tasks; and generating a sequence of causal improvement tasks for ineffective time-consuming tasks. The beneficial effects of this invention are: improving the accuracy of ineffective time analysis and improvement, and maximizing the efficiency of the hardware processing workshop.

Description

Technical Field

[0001] This invention relates to the field of big data management technology, specifically to a method for maximizing the efficiency of a hardware processing workshop based on big data. Background Technology

[0002] The current efficiency management methods in hardware processing workshops rely on subjective experience and judgment, resulting in inaccurate skill level assessments; time consumption analysis only stays at the overall task level and cannot accurately locate ineffective time-consuming links in specific processes; causal attribution is simplistic and crude, making it difficult to identify the root causes of multi-level deep problems; improvement measures are fragmented and priority settings lack data support; the static analysis models used cannot adapt to dynamic changes in the production line, ultimately leading to poor efficiency optimization and unreasonable resource allocation. Summary of the Invention

[0003] To address the aforementioned technical problems, this paper proposes a big data-based method for maximizing efficiency management in hardware processing workshops. This solution overcomes the shortcomings of existing efficiency management methods in hardware processing workshops, which rely on subjective experience for evaluation, leading to inaccurate skill level assessments; time-consuming analysis remains at the overall task level, failing to pinpoint ineffective time-consuming steps in specific processes; simplistic and crude causal attribution makes it difficult to identify multi-level, deep-seated problem roots; fragmented improvement measures lack data support for prioritization; and static analysis models are unable to adapt to dynamic changes in the production line, ultimately resulting in poor efficiency optimization and unreasonable resource allocation.

[0004] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0005] A big data-based method for maximizing efficiency management in a hardware processing workshop includes:

[0006] S1. Based on edge vision sensors, acquire video data of historical hardware workpiece processing tasks, analyze the proficiency of hardware workpiece processing tasks against a known personnel database, and quantify the proficiency categories of processing personnel in the hardware processing workshop.

[0007] S2. Cluster the processing workers in the hardware processing workshop according to their skill level, analyze the execution time of the processing actions corresponding to the hardware workpiece processing tasks in different skill level clusters, and initialize the time distribution of hardware workpiece processing actions under each skill level.

[0008] S3. Based on the time consumption distribution of hardware workpiece processing actions under different proficiency levels, construct a hardware workpiece processing action-effective / ineffective time consumption identification model, and mark the ineffective time consumption task type of hardware workpiece processing actions under the proficiency level of each processing worker;

[0009] S4. Based on the types of ineffective time-consuming tasks in hardware workpiece processing under the proficiency of each processing worker, trace the cause and effect of ineffective time-consuming tasks in hardware workpiece processing, build a cause and effect list of ineffective time-consuming tasks in hardware workpiece processing, and generate a sequence of cause and effect improvement tasks for ineffective time-consuming tasks in hardware workpiece processing.

[0010] Preferably, based on edge vision sensors, video data of hardware workpiece processing tasks are collected at several locations in the hardware processing workshop;

[0011] Based on a fixed frame rate interval, frame marking is performed on the video data of hardware workpiece processing task to obtain the key frame image video stream of hardware workpiece processing task.

[0012] Based on the ViBe algorithm, the pixel value neighbor variation value of the key frame image of the hardware workpiece processing task is recursively calculated according to the image timestamp of the video stream of the key frame image of the hardware workpiece processing task, and the foreground and background of the processing personnel in the hardware processing workshop in the key frame image of the hardware workpiece processing task are divided.

[0013] Based on the foreground of the processing personnel in the hardware processing workshop in the key frame image of the hardware workpiece processing task, the coordinate positions of the skeletal joints in the foreground of the processing personnel in the hardware processing workshop in the video stream of the key frame image of the hardware workpiece processing task are extracted using open source human vision detection.

[0014] Based on the coordinate positions of skeletal joints in the foreground of the processing personnel in the hardware processing workshop in the key frame image video stream of the hardware workpiece processing task, the unit timestamp is used as the observation window, and the coordinate positions of skeletal joints in the foreground of the processing personnel are used as the observation object to extract the temporal data of the processing action features of the processing personnel in the hardware processing workshop.

[0015] Based on the hardware workpiece processing task, several sets of standardized processing action video data of the processing personnel of the hardware workpiece processing task are pre-captured, and the time sequence data of the standardized processing action features of the processing personnel of the hardware workpiece processing task are extracted.

[0016] Based on the entire processing flow of hardware workpiece processing tasks, the processing action characteristics time sequence data of the processing personnel in the hardware processing workshop are divided to obtain an array of processing action characteristics time sequence data of the processing personnel in the hardware processing workshop.

[0017] Based on dynamic time warping, the spatial distance between the time sequence data array of processing action characteristics of processing personnel in the hardware processing workshop and the time sequence data of standardized processing action characteristics of processing personnel in hardware workpiece processing tasks is verified, and the deviation value of processing action characteristics of processing personnel in the hardware processing workshop is determined.

[0018] Preferably, the deviation values ​​of the processing action characteristics of the processing personnel in the hardware processing workshop are normalized, and the proportion of each processing action characteristic deviation value to the overall processing action characteristic deviation value is calculated.

[0019] Based on the proportion of each processing action feature deviation value to the overall processing action feature deviation value, calculate the processing action feature deviation information entropy of the processing personnel in the hardware processing workshop;

[0020] Based on the deviation of the processing action characteristics of the processing personnel in the hardware processing workshop from the information entropy, assign weights to the deviation of the processing action characteristics of the processing personnel in the hardware processing workshop.

[0021] Based on the weighted summation formula, the proficiency score of the processing personnel in the hardware processing workshop is calculated by using the deviation weights and deviation values ​​of the processing action characteristics of the processing personnel in the hardware processing workshop.

[0022] Preferably, based on preset proficiency tags, the proficiency scores of processing personnel in the hardware processing workshop are divided to determine the proficiency level of the processing personnel in the hardware processing workshop;

[0023] The skill level of the processing personnel in the hardware processing workshop is associated and bound with the time sequence data array of the processing action characteristics of the processing personnel in the hardware processing workshop, and a time sequence parameter matrix of processing action characteristics under the skill level of each processing personnel is established.

[0024] Based on the K-means clustering algorithm, the skill level of the processing personnel is used as the initial cluster, and each element in the temporal parameter matrix of the processing action features under each skill level is used as the clustering object to generate a set of temporal parameter clusters of processing action features under each skill level.

[0025] Using PCA principal component analysis, the dimensionality of the time series parameter clusters of processing action features under each proficiency level is reduced to obtain the dimensionality-reduced clusters of processing action feature time series parameters under each proficiency level.

[0026] Based on the entire processing flow of hardware workpiece processing tasks, the processing action feature time sequence parameter set under each proficiency level is divided according to each sub-processing flow, and the processing action feature time sequence parameter set under each proficiency level is obtained.

[0027] Mark the processing action feature time consumption sample of each sub-processing process in the set of processing action feature time parameter clusters for each sub-processing process under each proficiency level; the processing action feature time consumption sample includes: processing action type, start time and end time.

[0028] Based on the Weibull distribution, the probability distribution of the processing action feature time consumption samples of each sub-processing process in the set of time series parameters of processing action features of sub-processing processes under each skill level is calculated. The samples are then substituted into the maximum likelihood function estimation and fitting solution to obtain the processing action time distribution of hardware workpieces under each skill level.

[0029] Preferably, based on the processing time type of hardware workpieces, the processing time distribution of hardware workpieces under different skill levels is marked to obtain the effective / ineffective processing time feature data of hardware workpieces under different skill levels.

[0030] Based on the effective / ineffective time consumption characteristic data of hardware workpiece processing actions under various skill levels, a sliding window is used to statistically analyze the time series parameters of the effective / ineffective time consumption characteristic of hardware workpiece processing actions under various skill levels.

[0031] Establish a three-dimensional matrix of proficiency-material-time consumption, decompose the time distribution of hardware workpiece processing actions under each proficiency level, and obtain the structured parameters of hardware workpiece processing actions-effective / ineffective time consumption characteristics under each proficiency level;

[0032] Based on the Transformer architecture, the LTSM (Long Short-Term Memory) network is used as the first temporal feature decoder, and the random forest is used as the second structural feature decoder. The temporal parameters of the effective / ineffective time consumption features of hardware processing actions under each proficiency level and the structured parameters of the effective / ineffective time consumption features of hardware processing actions under each proficiency level are used as inputs. The trend vectors of the effective / ineffective time consumption features of hardware processing actions under each proficiency level and the contribution vectors of the structured features of the effective / ineffective time consumption features of hardware processing actions under each proficiency level are used as outputs.

[0033] Preferably, based on the entropy weight method, the information entropy of the trend vector of the effective / ineffective time consumption feature of the hardware workpiece processing action under each proficiency level per unit time is calculated;

[0034] Based on the profit distribution of cooperative game theory, the variance of the overall contribution value of the structured feature contribution vector of the processing actions of hardware workpieces with different proficiency levels per unit time is calculated.

[0035] Based on the information entropy of the time series parameters of the effective / ineffective time consumption characteristics of hardware workpiece processing actions under each proficiency level per unit time and the variance of the contribution value of the structured parameters of the effective / ineffective time consumption characteristics of hardware workpiece processing actions under each proficiency level, a dual-channel adaptive weighting mechanism is established, assigning weights to the trend vector of the effective / ineffective time consumption characteristics of hardware workpiece processing actions under each proficiency level and the contribution value vector of the structured parameters of the effective / ineffective time consumption characteristics of hardware workpiece processing actions under each proficiency level.

[0036] Based on the weighted concatenation method, according to the trend vector of effective / ineffective time consumption of hardware workpiece processing actions under each proficiency level and the weight of the structured feature contribution vector of effective / ineffective time consumption of hardware workpiece processing actions under each proficiency level, feature fusion is performed on the trend vector of effective / ineffective time consumption of hardware workpiece processing actions under each proficiency level and the weight of the structured feature contribution vector of effective / ineffective time consumption of hardware workpiece processing actions under each proficiency level. The result is substituted into the hardware workpiece processing time consumption type classification connection layer to generate the invalid time consumption task type of hardware workpiece processing actions under each proficiency level.

[0037] Preferably, based on the types of invalid time-consuming tasks in hardware workpiece processing under the proficiency of each processing worker, the nodes are the hardware workpiece processing action-effective / invalid time-consuming feature data and the structured parameters of the hardware workpiece processing action-effective / invalid time-consuming feature under the proficiency of each worker. The edges are the execution order of the entire processing process of the hardware workpiece processing task. The co-occurrence frequency of each node for the invalid time-consuming task type of hardware workpiece processing action is calculated and edge weights are assigned to establish an invalid time-consuming event graph.

[0038] Test independently according to the conditions, calculate the correlation between each node in the invalid time-consuming event graph, and determine the relative independent dependencies of each node in the invalid time-consuming event graph;

[0039] By using causal direction determination, the causal strength between each node in the invalid time-consuming event graph is calculated, and the relative connection causal relationship between each node in the invalid time-consuming event graph is determined.

[0040] Based on the relative independent dependencies and relative causal connections of each node in the invalid time-consuming event graph, the invalid time-consuming event graph is updated to obtain a causal directed graph of invalid time-consuming events.

[0041] Preferably, the FP-Growth frequent pattern mining algorithm is used to generate association rules for invalid time-consuming task types of hardware workpiece processing actions based on the relative independent dependencies and relative connection causal relationships of each node in the causal directed graph of invalid time-consuming events.

[0042] Based on the causal directed graph of invalid time-consuming events, extract the causal path of invalid time-consuming task type of hardware workpiece processing action under the proficiency of each processing worker;

[0043] By utilizing the association rules of invalid time-consuming task types in hardware workpiece processing actions, the causal path effect strength of invalid time-consuming task types in hardware workpiece processing actions is calculated under the proficiency of each processing worker.

[0044] Normalize the causal path effect intensity of the ineffective time-consuming task type of hardware workpiece processing action under the proficiency of each processing worker, assign weights to the corresponding causal nodes in the causal directed graph of ineffective time-consuming events, and obtain the causal weighted directed graph of ineffective time-consuming events.

[0045] Based on multi-factor linear regression, and according to the causal weighted directed graph of invalid time-consuming events, the invalid time-consuming task type of hardware workpiece processing action is calculated under the proficiency of each processing worker, and the causal priority of invalid time-consuming hardware workpiece processing action is determined.

[0046] Based on the causal priority of the ineffective time consumption of metal workpiece processing actions, a causal improvement task sequence for the ineffective time consumption of metal workpiece processing actions is constructed.

[0047] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0048] This invention proposes a big data-based efficiency maximization management scheme for hardware processing workshops. It constructs a dynamic evaluation system for worker proficiency based on edge vision and motion analysis, employs Weibull distribution and deep learning models to achieve refined modeling of process time consumption, identifies the root causes of ineffective work hours using causal reasoning graphs, and generates priority improvement sequences through adaptive optimization algorithms. The beneficial effects include: improving the depth of problem diagnosis in the processing process (positioning accuracy down to the second level) and the scientific nature of decision-making (enhancing the effectiveness of improvement measures), generating a closed-loop management scheme of monitoring-diagnosis-optimization, improving the overall processing efficiency of the workshop, and reducing quality defects. Attached Figure Description

[0049] Figure 1 This is a management solution for maximizing the efficiency of a hardware processing workshop based on big data. Detailed Implementation

[0050] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0051] Reference Figure 1 As shown, a big data-based method for maximizing efficiency management in a hardware processing workshop includes:

[0052] Step 1: Based on edge vision sensors, acquire video data of historical hardware workpiece processing tasks, analyze the proficiency of hardware workpiece processing tasks against a known personnel database, and quantify the proficiency categories of processing personnel in the hardware processing workshop.

[0053] Step one includes the following:

[0054] Based on edge vision sensors, video data of hardware workpiece processing tasks are collected at several locations in the hardware processing workshop.

[0055] Based on a fixed frame rate interval, frame marking is performed on the video data of hardware workpiece processing task to obtain the key frame image video stream of hardware workpiece processing task.

[0056] Based on the ViBe algorithm, the pixel value neighbor-to-neighbor variation of the keyframe image of the hardware processing task is recursively calculated according to the image timestamp of the video stream. This is used to divide the foreground and background of the processing personnel in the hardware processing workshop in the keyframe image of the hardware processing task, as follows:

[0057]

[0058] in, The foreground of the processing workers in the hardware processing workshop in the keyframe image of the hardware workpiece processing task at the t-th timestamp. The background of the processing workers in the hardware processing workshop is shown in the keyframe image of the hardware workpiece processing task at the t-th timestamp. Let be the pixel value in the keyframe image of the hardware workpiece processing task at the t-th timestamp. This represents the average background value in the keyframe image of the previous metal workpiece machining task. λ represents the background standard deviation in the keyframe image of the previous metal workpiece processing task, and λ represents the sensitivity of foreground detection.

[0059] Based on the foreground of the processing personnel in the hardware processing workshop in the key frame image of the hardware workpiece processing task, the coordinate positions of the skeletal joints in the foreground of the processing personnel in the hardware processing workshop in the video stream of the key frame image of the hardware workpiece processing task are extracted using open source human vision detection.

[0060] As a further point, the choice of framework for building open-source human visual detection includes, but is not limited to, OpenPose human pose recognition or MediaPipe, which are well known to those skilled in the art and will not be discussed in detail here.

[0061] Based on the coordinate positions of skeletal joints in the foreground of the processing personnel in the hardware processing workshop in the key frame image video stream of the hardware workpiece processing task, the unit timestamp is used as the observation window, and the coordinate positions of skeletal joints in the foreground of the processing personnel are used as the observation object to extract the temporal data of the processing action features of the processing personnel in the hardware processing workshop.

[0062] Based on the hardware workpiece processing task, several sets of standardized processing action video data of the processing personnel of the hardware workpiece processing task are pre-captured, and the time sequence data of the standardized processing action features of the processing personnel of the hardware workpiece processing task are extracted.

[0063] Based on the entire processing flow of hardware workpiece processing tasks, the processing action characteristics time sequence data of the processing personnel in the hardware processing workshop are divided to obtain an array of processing action characteristics time sequence data of the processing personnel in the hardware processing workshop.

[0064] Based on dynamic time warping, the spatial distance between the time series data array of processing action characteristics of processing personnel in the hardware processing workshop and the time series data of standardized processing action characteristics of processing personnel for hardware workpiece processing tasks is verified, and the deviation value of processing action characteristics of processing personnel in the hardware processing workshop is determined, as follows:

[0065]

[0066] in, Let G be the deviation value of the k-th machining action feature of the machining personnel in the hardware processing workshop, and DTW(G,Q) be the spatial distance between the machining action feature of the machining personnel in the hardware processing workshop and the standardized machining action feature of the machining personnel for the hardware workpiece machining task. k ξ represents the characteristic parameter of the kth processing action of the processing personnel in the hardware processing workshop, W represents the characteristic parameter of the kth standardized processing action of the processing personnel in the hardware workpiece processing task, and ξ represents the timing path alignment guide.

[0067] ---

[0068] The deviation values ​​of processing actions of processing personnel in the hardware processing workshop are normalized, and the proportion of each processing action deviation value to the overall processing action deviation value is calculated.

[0069] Based on the proportion of each processing action feature deviation value to the overall processing action feature deviation value, calculate the processing action feature deviation information entropy of the processing personnel in the hardware processing workshop;

[0070] Based on the deviation of the processing action characteristics of the processing personnel in the hardware processing workshop from the information entropy, assign weights to the deviation of the processing action characteristics of the processing personnel in the hardware processing workshop.

[0071] Based on the weighted summation formula, the proficiency score of the processing workers in the hardware processing workshop is calculated using the deviation weights and deviation values ​​of their processing action characteristics. The method is as follows:

[0072]

[0073] Among them, S v To rate the proficiency of the vth worker in the hardware processing workshop, W vk Let A be the feature deviation weight of the k-th processing action of the v-th worker in the hardware processing workshop. vkThis represents the deviation value of the k-th processing action characteristic of the v-th worker in the hardware processing workshop.

[0074] When using it, refer to the steps outlined above.

[0075] As a further point, the assessment of the proficiency of processing workers in traditional hardware processing workshops relies on subjective observation or simple time statistics, lacking objective data support. It only roughly counts task time or action classification, making it difficult to adapt to the changes in actions of different tasks, ignoring the continuity of time sequence, and because it lacks a refined comparison with standardized actions, it can only be indirectly assessed through output or completion time, resulting in a one-sided proficiency assessment that cannot provide accurate data support for improving overall production efficiency.

[0076] This solution uses edge vision sensors and the ViBe algorithm to quantify human movements in both the time and space dimensions. It combines dynamic time warping (DTW) to accurately compare the microscopic differences between actual and standardized movements, and automatically weights key deviations based on information entropy to ensure that proficiency scores are objective and accurate, providing dynamic data support for workshop efficiency optimization.

[0077] Step 2: Cluster the processing workers in the hardware processing workshop according to their skill level, analyze the execution time of the processing actions corresponding to the hardware workpiece processing tasks in different skill level clusters, and initialize the time distribution of hardware workpiece processing actions under each skill level.

[0078] Step two includes the following:

[0079] Based on preset proficiency tags, the proficiency scores of processing personnel in the hardware processing workshop are divided to determine the proficiency level of the processing personnel in the hardware processing workshop.

[0080] The skill level of the processing personnel in the hardware processing workshop is associated and bound with the time sequence data array of the processing action characteristics of the processing personnel in the hardware processing workshop, and a time sequence parameter matrix of processing action characteristics under the skill level of each processing personnel is established.

[0081] Based on the K-means clustering algorithm, the skill level of the processing personnel is used as the initial cluster, and each element in the temporal parameter matrix of the processing action features under each skill level is used as the clustering object to generate a set of temporal parameter clusters of processing action features under each skill level.

[0082] Using PCA principal component analysis, the dimensionality of the time series parameter clusters of processing action features under each proficiency level is reduced to obtain the dimensionality-reduced clusters of processing action feature time series parameters under each proficiency level.

[0083] Based on the entire processing flow of hardware workpiece processing tasks, the processing action feature time sequence parameter set under each proficiency level is divided according to each sub-processing flow, and the processing action feature time sequence parameter set under each proficiency level is obtained.

[0084] Mark the processing action feature time consumption sample of each sub-processing process in the set of processing action feature time parameter clusters for each sub-processing process under each proficiency level; the processing action feature time consumption sample includes: processing action type, start time and end time.

[0085] Based on the Weibull distribution, the probability distribution of the processing action feature time consumption samples of each sub-processing process in the set of time series parameters of processing action feature of sub-processing process under each proficiency level is calculated. The maximum likelihood function is substituted into the estimation and fitting solution to obtain the processing action time distribution of hardware workpieces under each proficiency level.

[0086] Furthermore, as an example: the detailed process for evaluating the time distribution of hardware workpiece processing actions at different skill levels is as follows:

[0087] 1. Proficiency level classification:

[0088] Set a scoring range (e.g., 0-100), dividing it into beginner (0-60), intermediate (60-85), and advanced (85-100), and bind the corresponding personnel's action sequence data.

[0089] 2. K-means clustering and PCA dimensionality reduction:

[0090] Input: Action sequence matrix of 100 people (dimension = 1000 × N, where N is the number of action features).

[0091] Procedure: First, divide the features into 3 clusters based on proficiency, then use PCA to reduce the features to 10 dimensions while retaining 95% of the variance.

[0092] Output: 3 sets of dimensionality-reduced clusters (e.g., the primary cluster contains 30 people, and the principal components of motion are concentrated in the grip stability dimension).

[0093] 3. Sub-process segmentation and time consumption marking:

[0094] The bearing turning task is divided into clamping → tool setting → rough turning → finish turning → inspection.

[0095] Sample of clamping time for junior personnel: Type = clamping, start time = 00:00, end time = 00:03:21.

[0096] 4. Weibull distribution fitting:

[0097] Input: 50 time samples (in seconds) of the clamping process for junior personnel.

[0098] Calculation: The shape parameter k = 1.2 and the scale parameter λ = 180 are obtained by maximum likelihood estimation, indicating that the time consumption of this step follows a right-skewed distribution.

[0099] Application: The 80% confidence interval for predicting when a junior operator completes the clamping process is [120s, 300s].

[0100] When using it, refer to the steps outlined above.

[0101] As a further point: Traditional methods mainly rely on human experience to classify proficiency levels, using static classification and simple time statistics, which cannot dynamically adapt to changes in different processing tasks; at the same time, they lack refined time consumption analysis of sub-processes, clustering methods do not consider the redundancy of high-dimensional time series data, and rely on simple models such as normal distribution, resulting in inaccurate evaluation results, poor adaptability, and difficulty in identifying specific inefficient links and the actual time consumption distribution.

[0102] This solution achieves objective and accurate proficiency classification through data-driven dynamic clustering and PCA dimensionality reduction, and accurately identifies inefficient links by combining refined sub-process analysis; it uses Weibull distribution to fit the time consumption characteristics of real processing scenarios to improve prediction accuracy; it achieves efficient data retrieval through matrix storage and forms an automated evaluation closed loop to eliminate human bias and accurately quantify the overall processing time distribution results in the processing workshop.

[0103] Step 3: Based on the time distribution of hardware workpiece processing actions under different proficiency levels, construct a hardware workpiece processing action-effective / ineffective time identification model, and mark the ineffective time-consuming task type of hardware workpiece processing actions under each processing worker's proficiency level;

[0104] Step three includes the following:

[0105] Based on the processing time type of hardware workpieces, the processing time distribution of hardware workpieces under different skill levels is labeled to obtain the effective / ineffective processing time feature data of hardware workpieces under different skill levels.

[0106] As a further point, the processing time types for the aforementioned hardware workpieces include: skill-based processing time (number of clamping and adjustment cycles), technical processing time (idle travel distance), and organizational processing time (equipment in standby mode).

[0107] Based on the effective / ineffective time consumption characteristic data of hardware workpiece processing actions under various skill levels, a sliding window is used to statistically analyze the time series parameters of the effective / ineffective time consumption characteristic of hardware workpiece processing actions under various skill levels.

[0108] As a further detail, the timing parameters of the effective / ineffective time consumption characteristics of the hardware workpiece processing action include: energy integral, zero crossing rate, and envelope area;

[0109] Establish a three-dimensional matrix of proficiency-material-time consumption, decompose the time distribution of hardware workpiece processing actions under each proficiency level, and obtain the structured parameters of hardware workpiece processing actions-effective / ineffective time consumption characteristics under each proficiency level;

[0110] As a further point, the structured parameters of the effective / ineffective time consumption characteristics of the hardware workpiece processing actions under the above-mentioned proficiency levels include: personnel attributes, material hardness, and cutting fluid temperature;

[0111] Based on the Transformer architecture, the LTSM (Long Short-Term Memory) network is used as the first temporal feature decoder, and the random forest is used as the second structural feature decoder. The temporal parameters of the effective / ineffective time consumption features of hardware processing actions under each proficiency level and the structured parameters of the effective / ineffective time consumption features of hardware processing actions under each proficiency level are used as inputs. The trend vector of the effective / ineffective time consumption features of hardware processing actions under each proficiency level and the contribution vector of the structured features of the effective / ineffective time consumption features of hardware processing actions under each proficiency level are used as outputs.

[0112] Based on the entropy weight method, the information entropy of the trend vector of the effective / ineffective time consumption feature of hardware workpiece processing actions under different skill levels per unit time is calculated.

[0113] Based on the profit distribution of cooperative game theory, the variance of the overall contribution value of the structured feature contribution vector of the processing actions of hardware workpieces with different proficiency levels per unit time is calculated.

[0114] Based on the information entropy of the time series parameters of the effective / ineffective time consumption characteristics of hardware workpiece processing actions under each proficiency level per unit time and the variance of the contribution value of the structured parameters of the effective / ineffective time consumption characteristics of hardware workpiece processing actions under each proficiency level, a dual-channel adaptive weighting mechanism is established, assigning weights to the trend vector of the effective / ineffective time consumption characteristics of hardware workpiece processing actions under each proficiency level and the contribution value vector of the structured parameters of the effective / ineffective time consumption characteristics of hardware workpiece processing actions under each proficiency level.

[0115] Based on the weighted concatenation method, according to the trend vector of effective / ineffective time consumption of hardware workpiece processing actions under each proficiency level and the weight of the structured feature contribution vector of effective / ineffective time consumption of hardware workpiece processing actions under each proficiency level, feature fusion is performed on the trend vector of effective / ineffective time consumption of hardware workpiece processing actions under each proficiency level and the weight of the structured feature contribution vector of effective / ineffective time consumption of hardware workpiece processing actions under each proficiency level. The result is substituted into the hardware workpiece processing time consumption type classification connection layer to generate the invalid time consumption task type of hardware workpiece processing actions under each proficiency level.

[0116] Furthermore, for example: the specific types of ineffective time-consuming tasks in the processing of hardware workpieces under different operator skill levels are as follows:

[0117] LSTM timing decoder output:

[0118] Trend feature vectors include: spindle current fluctuation trend intensity (0.12 indicates stable), and probability of sudden abnormal vibration (0.83 indicates high risk).

[0119] Random Forest Output:

[0120] Structured contribution vectors include: skill level (0.32), number of times the workpiece has been machined (0.18), material hardness (0.12), tool wear (0.9), depth of cut (0.5), and cutting fluid temperature (0.24).

[0121] Feature fusion and weighted splicing:

[0122] Trend vector weight: 0.7, Contribution vector weight: 0.3

[0123] Fusion vector: [0.12*0.7, -0.83*0.7, ..., 0.32*0.3, 0.18*0.3, ...]

[0124] When using it, refer to the steps outlined above.

[0125] As a further point, existing methods can only roughly determine the overall task timeout in time consumption analysis but cannot accurately locate specific inefficient links (such as clamping, debugging, etc.). Furthermore, they simply attribute abnormal time consumption to equipment failure while ignoring personnel operation and process factors. The use of static threshold judgment makes it impossible to distinguish the reasonable differences in operation between personnel with different skill levels. The modeling method is too simple, and the feature weights rely on manual setting and lack the ability to adapt to new types of workpieces, ultimately resulting in inaccurate time consumption analysis.

[0126] This solution is based on a multimodal architecture, utilizing LSTM + Random Forest to capture anomalies in operational time sequences and analyze the structural impact of process parameters. Furthermore, through an adaptive weight allocation mechanism optimized by entropy weighting and game theory, feature weights can be automatically adjusted based on processing time parameters and different processing materials. Ultimately, this approach enables precise location of time-consuming anomalies in specific processes, from anomaly detection to root cause analysis.

[0127] Step 4: Based on the types of ineffective time-consuming tasks in hardware workpiece processing under the proficiency of each processing worker, trace the cause and effect of ineffective time-consuming tasks in hardware workpiece processing, build a cause and effect list of ineffective time-consuming tasks in hardware workpiece processing, and generate a sequence of cause and effect improvement tasks for ineffective time-consuming tasks in hardware workpiece processing.

[0128] Step four includes the following:

[0129] Based on the types of invalid time-consuming tasks in hardware workpiece processing under the proficiency of each processing worker, the structured parameters of the effective / invalid time-consuming features of hardware workpiece processing actions under each proficiency level are used as nodes, and the execution order of the entire processing flow of hardware workpiece processing tasks is used as edges. The co-occurrence frequency of each node for the invalid time-consuming task type of hardware workpiece processing actions is calculated and edge weights are assigned to establish an invalid time-consuming event graph.

[0130] Based on independent testing according to conditions, calculate the correlation between each node in the invalid time-consuming event graph, and determine the relative independent dependencies of each node in the invalid time-consuming event graph, as follows:

[0131]

[0132] Where p-value represents the saliency probability between nodes in the invalid time-consuming event graph, χ² 2 Let ρ be the chi-square distribution function, m be the total number of nodes in the invalid time-consuming event graph, and ρ be the total number of nodes in the graph. 2 XY|Z Let z be the partial correlation coefficient between node X and node Y given variable z.

[0133] By utilizing causal direction determination, the causal strength between each node in the invalid time-consuming event graph is calculated, and the relative connection causal relationships between each node in the invalid time-consuming event graph are determined, as follows:

[0134]

[0135] Among them, F X→Y To calculate the causal influence strength of node X on node Y in the invalid time-consuming event graph, Var(∈Y,t) is the residual variance when predicting given the data of node Y at the t-th unit time, Var(∈Y|X,t) is the residual variance when predicting node Y given the data of node X at the t-th unit time, and ln is the natural logarithm.

[0136] Based on the relative independent dependencies and relative connection causal relationships of each node in the invalid time-consuming event graph, the invalid time-consuming event graph is updated to obtain a causal directed graph of invalid time-consuming events.

[0137] Using the FP-Growth frequent pattern mining algorithm, association rules for invalid time-consuming task types of hardware workpiece processing actions are generated based on the relative independent dependencies and relative connection causal relationships of each node in the causal directed graph of invalid time-consuming events.

[0138] Based on the causal directed graph of invalid time-consuming events, extract the causal path of invalid time-consuming task type of hardware workpiece processing action under the proficiency of each processing worker;

[0139] Using the association rules for ineffective time-consuming tasks in hardware workpiece processing, the causal path effect strength of ineffective time-consuming tasks in hardware workpiece processing is calculated for each worker's skill level, as follows:

[0140]

[0141] Among them, V X→Y Let conf(X→Y) represent the causal path effect strength of the ineffective time-consuming task type in hardware workpiece processing under different operator skill levels, and let conf(X→Y) represent the confidence of the association rule for the ineffective time-consuming task type in hardware workpiece processing. The confidence level of the association rule is set so that no invalid time-consuming task type of hardware workpiece processing action occurs.

[0142] Normalize the causal path effect intensity of the ineffective time-consuming task type of hardware workpiece processing action under the proficiency of each processing worker, assign weights to the corresponding causal nodes in the causal directed graph of ineffective time-consuming events, and obtain the causal weighted directed graph of ineffective time-consuming events.

[0143] Based on multi-factor linear regression, and according to the causal weighted directed graph of invalid time-consuming events, the invalid time-consuming task type of hardware workpiece processing action is calculated under the proficiency of each processing worker, and the causal priority of invalid time-consuming hardware workpiece processing action is determined.

[0144] Based on the causal priority of the ineffective time consumption of metal workpiece processing actions, a causal improvement task sequence for the ineffective time consumption of metal workpiece processing actions is constructed.

[0145] When using it, refer to the steps outlined above.

[0146] As a further point, existing methods, in analyzing ineffective time consumption, only focus on superficial phenomena (such as timeout-equipment failure), heavily relying on subjective experience and lacking data support; improvement suggestions are fragmented and fail to identify the chain effects between processes; task priority setting relies on human experience, leading to resource misallocation; and the static association rule mining algorithm used cannot adapt to dynamic changes in the production line, causing improvement measures to fail in new process scenarios.

[0147] This solution is based on event graph technology to trace multi-level causal chains (such as tracing from operational problems to training system defects), uses the FP-Growth algorithm to identify cross-process correlations, automatically quantifies the contribution of each factor through dynamic weight optimization, and combines multi-factor regression to achieve intelligent sorting of improvement tasks, enabling adaptive updating capabilities. This ensures that improvement measures are always synchronized with actual changes on the production line, improves the accuracy of ineffective time analysis and improvement, and maximizes the efficiency of the metal processing workshop.

[0148] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. A method for maximizing efficiency management in a hardware processing workshop based on big data, characterized in that, include: S1. Based on edge vision sensors, acquire video data of historical hardware workpiece processing tasks, analyze the proficiency of hardware workpiece processing tasks against a known personnel database, and quantify the proficiency categories of processing personnel in the hardware processing workshop. S2. Cluster the processing workers in the hardware processing workshop according to their skill level, analyze the execution time of the processing actions corresponding to the hardware workpiece processing tasks in different skill level clusters, and initialize the time distribution of hardware workpiece processing actions under each skill level. S3. Based on the time distribution of hardware workpiece processing actions at various skill levels, construct a hardware workpiece processing action-effective / ineffective time identification model to mark the types of ineffective time-consuming tasks for hardware workpiece processing actions at each worker's skill level, including: Based on the processing time type of hardware workpieces, the processing time distribution of hardware workpieces under different skill levels is labeled to obtain the effective / ineffective processing time feature data of hardware workpieces under different skill levels. Based on the effective / ineffective time consumption characteristic data of hardware workpiece processing actions under various skill levels, a sliding window is used to statistically analyze the time series parameters of the effective / ineffective time consumption characteristic of hardware workpiece processing actions under various skill levels. Establish a three-dimensional matrix of proficiency-material-time consumption, decompose the time distribution of hardware workpiece processing actions under each proficiency level, and obtain the structured parameters of hardware workpiece processing actions-effective / ineffective time consumption characteristics under each proficiency level; Based on the Transformer architecture, the LTSM (Long Short-Term Memory) network is used as the first temporal feature decoder, and the random forest is used as the second structural feature decoder. The temporal parameters of the effective / ineffective time consumption features of hardware processing actions under each proficiency level and the structured parameters of the effective / ineffective time consumption features of hardware processing actions under each proficiency level are used as inputs. The trend vector of the effective / ineffective time consumption features of hardware processing actions under each proficiency level and the contribution vector of the structured features of the effective / ineffective time consumption features of hardware processing actions under each proficiency level are used as outputs. Based on the entropy weight method, the information entropy of the trend vector of the effective / ineffective time consumption feature of the hardware workpiece processing action under each proficiency level per unit time is calculated. Based on the profit distribution of cooperative game theory, the variance of the overall contribution value of the structured feature contribution vector of the processing actions of hardware workpieces with different proficiency levels per unit time is calculated. Based on the information entropy of the time series parameters of the effective / ineffective time consumption characteristics of hardware workpiece processing actions under each proficiency level per unit time and the variance of the contribution value of the structured parameters of the effective / ineffective time consumption characteristics of hardware workpiece processing actions under each proficiency level, a dual-channel adaptive weighting mechanism is established, assigning weights to the trend vector of the effective / ineffective time consumption characteristics of hardware workpiece processing actions under each proficiency level and the contribution value vector of the structured parameters of the effective / ineffective time consumption characteristics of hardware workpiece processing actions under each proficiency level. Based on the weighted splicing method, according to the trend vector of effective / ineffective time consumption of hardware workpiece processing actions under each proficiency level and the weight of the structured feature contribution vector of effective / ineffective time consumption of hardware workpiece processing actions under each proficiency level, feature fusion is performed on the trend vector of effective / ineffective time consumption of hardware workpiece processing actions under each proficiency level and the weight of the structured feature contribution vector of effective / ineffective time consumption of hardware workpiece processing actions under each proficiency level. The result is then substituted into the hardware workpiece processing time consumption type classification connection layer to generate the ineffective time consumption task type of hardware workpiece processing actions under each proficiency level. S4. Based on the types of ineffective time-consuming tasks in hardware workpiece processing under the proficiency of each processing worker, trace the cause and effect of ineffective time-consuming tasks in hardware workpiece processing, build a cause and effect list of ineffective time-consuming tasks in hardware workpiece processing, and generate a sequence of cause and effect improvement tasks for ineffective time-consuming tasks in hardware workpiece processing.

2. The method for maximizing efficiency management in a hardware processing workshop based on big data, as described in claim 1, is characterized in that... S1 includes: Based on edge vision sensors, video data of hardware workpiece processing tasks are collected at several locations in the hardware processing workshop. Based on a fixed frame rate interval, frame marking is performed on the video data of hardware workpiece processing task to obtain the key frame image video stream of hardware workpiece processing task. Based on the ViBe algorithm, the pixel value neighbor variation value of the key frame image of the hardware workpiece processing task is recursively calculated according to the image timestamp of the video stream of the key frame image of the hardware workpiece processing task, and the foreground and background of the processing personnel in the hardware processing workshop in the key frame image of the hardware workpiece processing task are divided. Based on the foreground of the processing personnel in the hardware processing workshop in the key frame image of the hardware workpiece processing task, the coordinate positions of the skeletal joints in the foreground of the processing personnel in the hardware processing workshop in the video stream of the key frame image of the hardware workpiece processing task are extracted using open source human vision detection. Based on the coordinate positions of skeletal joints in the foreground of the processing personnel in the hardware processing workshop in the key frame image video stream of the hardware workpiece processing task, the unit timestamp is used as the observation window, and the coordinate positions of skeletal joints in the foreground of the processing personnel are used as the observation object to extract the temporal data of the processing action features of the processing personnel in the hardware processing workshop. Based on the hardware workpiece processing task, several sets of standardized processing action video data of the processing personnel of the hardware workpiece processing task are pre-captured, and the time sequence data of the standardized processing action features of the processing personnel of the hardware workpiece processing task are extracted. Based on the entire processing flow of hardware workpiece processing tasks, the processing action characteristics time sequence data of the processing personnel in the hardware processing workshop are divided to obtain an array of processing action characteristics time sequence data of the processing personnel in the hardware processing workshop. Based on dynamic time warping, the spatial distance between the time sequence data array of processing action characteristics of processing personnel in the hardware processing workshop and the time sequence data of standardized processing action characteristics of processing personnel in hardware workpiece processing tasks is verified, and the deviation value of processing action characteristics of processing personnel in the hardware processing workshop is determined.

3. The method for maximizing efficiency management in a hardware processing workshop based on big data, as described in claim 2, is characterized in that... S1 further includes: The deviation values ​​of processing actions of processing personnel in the hardware processing workshop are normalized, and the proportion of each processing action deviation value to the overall processing action deviation value is calculated. Based on the proportion of each processing action feature deviation value to the overall processing action feature deviation value, calculate the processing action feature deviation information entropy of the processing personnel in the hardware processing workshop; Based on the deviation of the processing action characteristics of the processing personnel in the hardware processing workshop from the information entropy, assign weights to the deviation of the processing action characteristics of the processing personnel in the hardware processing workshop. Based on the weighted summation formula, the proficiency score of the processing personnel in the hardware processing workshop is calculated by using the deviation weights and deviation values ​​of the processing action characteristics of the processing personnel in the hardware processing workshop.

4. The method for maximizing efficiency management in a hardware processing workshop based on big data, as described in claim 3, is characterized in that... S2 includes: Based on preset proficiency tags, the proficiency scores of processing personnel in the hardware processing workshop are divided to determine the proficiency level of the processing personnel in the hardware processing workshop. The skill level of the processing personnel in the hardware processing workshop is associated and bound with the time sequence data array of the processing action characteristics of the processing personnel in the hardware processing workshop, and a time sequence parameter matrix of processing action characteristics under the skill level of each processing personnel is established. Based on the K-means clustering algorithm, the skill level of the processing personnel is used as the initial cluster, and each element in the temporal parameter matrix of the processing action features under each skill level is used as the clustering object to generate a set of temporal parameter clusters of processing action features under each skill level. Using PCA principal component analysis, the dimensionality of the time series parameter clusters of processing action features under each proficiency level is reduced to obtain the dimensionality-reduced clusters of processing action feature time series parameters under each proficiency level. Based on the entire processing flow of hardware workpiece processing tasks, the processing action feature time sequence parameter set under each proficiency level is divided according to each sub-processing flow, and the processing action feature time sequence parameter set under each proficiency level is obtained. Mark the processing action feature time consumption sample of each sub-processing process in the set of processing action feature time parameter clusters for each sub-processing process under each proficiency level; the processing action feature time consumption sample includes: processing action type, start time and end time. Based on the Weibull distribution, the probability distribution of the processing action feature time consumption samples of each sub-processing process in the set of time series parameters of processing action features of sub-processing processes under each skill level is calculated. The samples are then substituted into the maximum likelihood function estimation and fitting solution to obtain the processing action time distribution of hardware workpieces under each skill level.

5. The method for maximizing efficiency management in a hardware processing workshop based on big data, as described in claim 4, is characterized in that... S4 includes: Based on the types of invalid time-consuming tasks in hardware workpiece processing under the proficiency of each processing worker, the structured parameters of the effective / invalid time-consuming features of hardware workpiece processing actions under each proficiency level are used as nodes, and the execution order of the entire processing flow of hardware workpiece processing tasks is used as edges. The co-occurrence frequency of each node for the invalid time-consuming task type of hardware workpiece processing actions is calculated and edge weights are assigned to establish an invalid time-consuming event graph. Test independently according to the conditions, calculate the correlation between each node in the invalid time-consuming event graph, and determine the relative independent dependencies of each node in the invalid time-consuming event graph; By using causal direction determination, the causal strength between each node in the invalid time-consuming event graph is calculated, and the relative connection causal relationship between each node in the invalid time-consuming event graph is determined. Based on the relative independent dependencies and relative causal connections of each node in the invalid time-consuming event graph, the invalid time-consuming event graph is updated to obtain a causal directed graph of invalid time-consuming events.

6. The method for maximizing efficiency management in a hardware processing workshop based on big data, as described in claim 5, is characterized in that... S4 further includes: Using the FP-Growth frequent pattern mining algorithm, association rules for invalid time-consuming task types of hardware workpiece processing actions are generated based on the relative independent dependencies and relative connection causal relationships of each node in the causal directed graph of invalid time-consuming events. Based on the causal directed graph of invalid time-consuming events, extract the causal path of invalid time-consuming task type of hardware workpiece processing action under the proficiency of each processing worker; By utilizing the association rules of invalid time-consuming task types in hardware workpiece processing actions, the causal path effect strength of invalid time-consuming task types in hardware workpiece processing actions is calculated under the proficiency of each processing worker. Normalize the causal path effect intensity of the ineffective time-consuming task type of hardware workpiece processing action under the proficiency of each processing worker, assign weights to the corresponding causal nodes in the causal directed graph of ineffective time-consuming events, and obtain the causal weighted directed graph of ineffective time-consuming events. Based on multi-factor linear regression, and according to the causal weighted directed graph of invalid time-consuming events, the invalid time-consuming task type of hardware workpiece processing action is calculated under the proficiency of each processing worker, and the causal priority of invalid time-consuming hardware workpiece processing action is determined. Based on the causal priority of the ineffective time consumption of metal workpiece processing actions, a causal improvement task sequence for the ineffective time consumption of metal workpiece processing actions is constructed.

Images