Factory production quality detection method and system based on Internet of Things

Abstract

The invention provides a factory production quality detection method and system based on the Internet of Things, and the method comprises the steps: carrying out the regional semantic perception association of sensor combinations in different regions, and carrying out the three-dimensional association processing of physical data through a self-adaptive transmission protocol, and obtaining a structured quality data flow; according to the structured quality data flow and the quality inspection parameters, performing differential quality inspection load balanced distribution and multi-stop collaborative reasoning on the logistics stop network to obtain semi-structured quality inspection result data; according to the quality inspection result data, topological path quality decline analysis and quality anomaly spatial-temporal clustering processing are carried out on the goods allocation and the product quality state, and a quality knowledge graph is obtained; and according to the quality knowledge graph, performing multi-dimensional quality control limit self-learning adjustment and quality early warning grading response processing on the quality inspection parameters of different goods locations and paths. According to the method, accurate traceability and rapid improvement of product quality problems are realized, and the efficiency and accuracy of quality detection are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of Internet of Things, and in particular to a factory production quality detection method and system based on the Internet of Things. Background Art

[0002] The application of IoT technology in the field of factory production quality control is becoming more and more widespread. Traditional quality inspection methods are mainly based on sampling inspection after the product is offline or data analysis of individual fixed inspection points on the production line. This method can no longer meet the needs of enterprises for full-process quality control in the intelligent manufacturing environment.

[0003] In recent years, the IoT-based quality inspection system has achieved real-time data collection and preliminary analysis during the production process by deploying a large number of sensors in the factory, and has improved the detection rate of quality problems to a certain extent. Such systems usually include four parts: data collection layer, transmission layer, processing layer and application layer, which can monitor the quality status of products at specific inspection points.

[0004] However, the existing IoT-based quality inspection system still has a key technical problem: it is unable to effectively analyze and track the quality changes of products as they move between different areas and logistics routes within the factory, making it difficult to determine the source and root causes of quality problems. Summary of the invention

[0005] The main purpose of the present invention is to solve the technical problem that the existing factory production quality inspection system based on the Internet of Things cannot effectively analyze and track the quality change rules of products during the movement between different areas and logistics paths in the factory, resulting in difficulty in determining the generation links and root causes of quality problems; A first aspect of the present invention provides a factory production quality detection method based on the Internet of Things, and the factory production quality detection method based on the Internet of Things includes: According to the regional division in the factory area modeling, the sensor combinations in different areas of the factory are associated with regional semantic perception, and the collected physical data is processed in three dimensions of area-equipment-data through the preset adaptive transmission protocol to obtain structured quality data stream; According to the structured quality data stream and preset quality inspection parameters, differential quality inspection load balancing allocation and multi-stop collaborative reasoning processing are performed on the logistics stop network in the factory to obtain semi-structured quality inspection result data containing defect type, location and severity information; According to the semi-structured quality inspection result data, topological path quality decline analysis and quality anomaly spatiotemporal clustering processing are performed on the cargo locations and product quality status in the factory to obtain the corresponding quality knowledge graph; According to the quality knowledge graph, multi-dimensional quality control limit self-learning adjustment and quality warning graded response processing are performed on the quality inspection parameters of different cargo locations and routes.

[0006] Optionally, in a first implementation of the first aspect of the present invention, according to the regional division in the factory area modeling, the sensor combinations in different areas of the factory are associated with regional semantic perception, and the collected physical data are processed with region-equipment-data three-dimensional association through a preset adaptive transmission protocol to obtain a structured quality data stream including: Semantically label the region according to the production function attributes of the factory region, extract the production stage, product type and quality requirement characteristics of the region, and construct the regional semantic feature vector; Mapping and associating the regional semantic feature vector with the existing sensor network in the region, assigning a regional semantic identity to each sensor, establishing a binding relationship between sensor data and regional semantics, and obtaining a semantically aware sensor data collection strategy; According to the semantically-aware sensor data collection strategy and the real-time production rhythm of each area, a preset adaptive transmission protocol is activated to schedule the transmission of the physical data collected by the sensor, so as to obtain a transmission-optimized original data set; The original data set is subjected to region-device-data three-dimensional association processing, and the data is associated and labeled with corresponding region semantic attributes, sensor characteristics, and data acquisition background to obtain a structured quality data stream.

[0007] Optionally, in a second implementation of the first aspect of the present invention, the differential quality inspection load balancing allocation and multi-stop collaborative reasoning processing are performed on the logistics stop network in the factory according to the structured quality data stream and the preset quality inspection parameters, and the semi-structured quality inspection result data containing defect type, location and severity information is obtained, including: Match the preset quality inspection parameters with the physical location and processing capacity of each logistics stop in the factory's logistics stop network to build a parameterized stop resource distribution matrix; According to the stop point resource distribution matrix and the structured quality data flow, the quality inspection task allocation plan of each stop point is calculated and dynamically adjusted to achieve differential quality inspection load balancing distribution and obtain a load-balanced dynamic task allocation plan; According to the load-balanced dynamic task allocation scheme and preset quality inspection parameters, the quality inspection algorithm is executed in parallel at each stop, and multi-stop collaborative reasoning is performed for complex defects to obtain a decentralized preliminary quality inspection result set; The scattered preliminary quality inspection result sets are aggregated, fused and checked for consistency to form semi-structured quality inspection result data containing defect type, location and severity information.

[0008] Optionally, in a third implementation of the first aspect of the present invention, the quality inspection algorithm is executed in parallel at each stop point according to the dynamic task allocation scheme of the load balancing and the preset quality inspection parameters, and multi-stop collaborative reasoning is performed for complex defects, and the decentralized preliminary quality inspection result set is obtained, including: According to the load-balanced dynamic task allocation scheme, the structured quality data stream is divided into multiple data subsets according to spatial distribution and data characteristics, and each data subset is distributed to the corresponding logistics stop point to obtain a stop point level quality inspection data allocation scheme; According to the quality inspection data allocation scheme and the preset quality inspection parameters, feature extraction and defect pattern matching processing are performed on the allocated data subset at each stop point, abnormal patterns are identified and the position and size of defects are calculated to obtain a defect candidate set at the stop point level; For complex defects in the defect candidate set, a multi-stop collaborative determination request is constructed, defect-related data is sent to adjacent stops, multi-source data fusion and arbitration processing are performed, and a collaboratively verified defect conclusion is obtained; The defect candidate sets and defect conclusions of each stop point are integrated into a quality inspection report in a unified format to obtain a decentralized preliminary quality inspection result set.

[0009] Optionally, in a fourth implementation of the first aspect of the present invention, the topological path quality decline analysis and quality anomaly spatiotemporal clustering processing are performed on the cargo locations and product quality status in the factory according to the semi-structured quality inspection result data to obtain the corresponding quality knowledge graph, including: According to the layout of the factory's logistics system and semi-structured quality inspection result data, a topological connection relationship model of the cargo locations in the factory is constructed; According to the topological connection relationship model and semi-structured quality inspection result data, the quality impact coefficient of each cargo location is calculated to obtain the weighted cargo location quality impact network; By using the cargo location quality influence network, the quality change gradient of the product between the cargo locations is calculated and the path segments where the quality drops sharply are detected, and a topological path quality decrease analysis is performed to obtain a set of quality critical paths; The quality critical path set is clustered according to the time and space dimensions to identify quality anomaly groups with similar time and space characteristics. In combination with the cargo location quality impact network, the source node and diffusion path model of the quality problem are constructed to obtain a quality knowledge graph.

[0010] Optionally, in a fifth implementation of the first aspect of the present invention, the use of the cargo location quality influence network to calculate the quality change gradient of the product between the cargo locations and detect the path segments where the quality drops sharply, and to perform topological path quality decrease analysis to obtain a quality critical path set includes: According to the product identification information and detection timestamp in the semi-structured quality inspection result data, the movement trajectory of the product in the storage location network is reconstructed to obtain the three-dimensional correlation sequence of product-storage location-time; According to the three-dimensional association sequence and the cargo location quality influence network, the change range of each key quality indicator of the product during the movement between adjacent cargo locations is calculated, a quality change vector with a cargo location weight influence factor is constructed, and a quality change gradient set at the path segment level is obtained; Applying the weight relationship in the cargo location quality impact network to the quality change gradient set, executing a weighted anomaly detection algorithm, and identifying a path segment whose quality change rate exceeds a weighted threshold; The path segments are aggregated and analyzed according to their connection relationships in the cargo location quality influence network to obtain path combinations, and the cumulative quality risk scores of each path combination are calculated to obtain a quality critical path set.

[0011] Optionally, in a sixth implementation of the first aspect of the present invention, performing multi-dimensional quality control limit self-learning adjustment and quality warning graded response processing on quality inspection parameters of different cargo locations and paths according to the quality knowledge graph includes: According to the source nodes and diffusion path model in the quality knowledge graph, the quality risk level of each cargo location in the factory is classified to obtain the cargo location risk level distribution map; According to the cargo location risk level distribution map, performing differentiated configuration on quality inspection parameters in the logistics stop point network related to each cargo location to obtain a differentiated quality inspection parameter set; Based on the differential quality inspection parameter set, combined with historical quality inspection data and actual quality results, the strength of association between quality inspection parameters and quality indicators is calculated using a preset parameter impact assessment algorithm to obtain a quality inspection parameter impact weight matrix; According to the quality inspection parameter influence weight matrix, the upper and lower control limits of the quality inspection parameters are automatically adjusted to obtain an adaptive quality inspection control limit set, and quality warning graded response processing is performed according to the adaptive quality inspection control limit set and the diffusion path model of the quality knowledge graph.

[0012] A second aspect of the present invention provides a factory production quality detection system based on the Internet of Things, and the factory production quality detection system based on the Internet of Things includes: The semantic association module is used to perform regional semantic perception association on sensor combinations in different areas of the factory according to the regional division in the factory area modeling, and perform region-equipment-data three-dimensional association processing on the collected physical data through a preset adaptive transmission protocol to obtain a structured quality data stream; A collaborative inspection module is used to perform differential quality inspection load balancing distribution and multi-stop collaborative reasoning processing on the logistics stop network in the factory according to the structured quality data stream and preset quality inspection parameters, and obtain semi-structured quality inspection result data containing defect type, location and severity information; A path analysis module is used to perform topological path quality decline analysis and quality anomaly spatiotemporal clustering processing on the cargo locations and product quality status in the factory according to the semi-structured quality inspection result data, so as to obtain a corresponding quality knowledge graph; The parameter optimization module is used to perform multi-dimensional quality control limit self-learning adjustment and quality warning graded response processing on the quality inspection parameters of different cargo locations and routes according to the quality knowledge graph.

[0013] The above-mentioned factory production quality inspection method and system based on the Internet of Things, through regional semantic perception association of sensor combinations in different areas, and through adaptive transmission protocol, performs three-dimensional association processing on physical data to obtain structured quality data stream; according to the structured quality data stream and quality inspection parameters, differential quality inspection load balancing distribution and multi-stop collaborative reasoning are performed on the logistics stop network to obtain semi-structured quality inspection result data; according to the quality inspection result data, topological path quality reduction analysis and quality abnormality spatiotemporal clustering processing are performed on the cargo location and product quality status to obtain the quality knowledge graph; according to the quality knowledge graph, multi-dimensional quality control limit self-learning adjustment and quality warning graded response processing are performed on the quality inspection parameters of different cargo locations and paths. This method realizes the accurate traceability and rapid improvement of product quality problems, and improves the efficiency and accuracy of quality inspection.

[0014] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description, claims and drawings.

[0015] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a schematic diagram of a first embodiment of a factory production quality detection method based on the Internet of Things in an embodiment of the present invention; Figure 2 Schematic diagram of an embodiment of a factory production quality inspection system based on the Internet of Things in an embodiment of the present invention. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0018] The terms "including" and "having" and any variations thereof mentioned in the embodiments of the present invention are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device end including a series of steps or units is not limited to the listed steps or units, but may optionally include other steps or units that are not listed, or may optionally include other steps or units that are inherent to these processes, methods, products or device ends.

[0019] To facilitate understanding of this embodiment, firstly, a factory production quality detection method based on the Internet of Things disclosed in an embodiment of the present invention is described in detail. Figure 1 As shown, the method comprises the following steps: 101. According to the regional division in the factory area modeling, the sensor combinations in different areas of the factory are associated with regional semantic perception, and the collected physical data are processed in three dimensions of area-equipment-data through the preset adaptive transmission protocol to obtain structured quality data stream; In one embodiment of the present invention, the sensor combinations in different areas of the factory are subjected to regional semantic perception association according to the regional division in the factory area modeling, and the collected physical data are subjected to region-device-data three-dimensional association processing through a preset adaptive transmission protocol to obtain a structured quality data stream, including: semantically labeling the area according to the production function attributes of the factory area, extracting the production stage, product type and quality requirement characteristics of the area, and constructing a regional semantic feature vector; mapping and associating the regional semantic feature vector with the existing sensor network in the area, assigning a regional semantic identity to each sensor, establishing a binding relationship between sensor data and regional semantics, and obtaining a semantically aware sensor data acquisition strategy; according to the semantically aware sensor data acquisition strategy and the real-time production rhythm of each area, activating the preset adaptive transmission protocol to schedule the transmission of the physical data collected by the sensor, and obtaining a transmission-optimized original data set; performing region-device-data three-dimensional association processing on the original data set, and associating and marking the data with the corresponding regional semantic attributes, sensor characteristics and data acquisition background, and obtaining a structured quality data stream.

[0020] Specifically, according to the regional division in the factory area modeling, the process of regional semantic perception association of sensor combinations in different areas of the factory first requires semantic labeling of the production function attributes of the factory area. This process constructs a regional semantic feature vector by extracting the production stage, product type and quality requirement characteristics of the area. For example, in a home appliance manufacturing factory, the assembly area is marked as "back-end assembly-refrigerator host-high-precision assembly requirements", the injection molding area is marked as "front-end molding-plastic shell-surface finish requirements", and the test area is marked as "final inspection-whole machine-functional integrity requirements". This semantic labeling digitizes the functional attributes of the area into a multi-dimensional vector, and each dimension corresponds to a semantic feature, such as the timing value of the production stage, the coding value of the product type, the strictness value of the quality requirements, etc., thereby forming a semantic feature vector of the area.

[0021] Specifically, mapping and associating the regional semantic feature vector with the existing sensor network in the region is a key step in achieving semantic awareness. This step assigns a regional semantic identity to each sensor, establishes a binding relationship between sensor data and regional semantics, and thus obtains a semantically aware sensor data acquisition strategy. In actual implementation, the vibration sensor in the assembly area is assigned a semantic identity of "back-end assembly-equipment vibration monitoring-precision assembly quality influencing factors", and the temperature sensor in the injection molding area is assigned a semantic identity of "front-end molding-ambient temperature monitoring-surface quality influencing factors". This association enables the sensor to not only know the physical quantity it monitors, but also the relationship between the physical quantity and the production semantics of the region, which enhances the contextual meaning of the data. The semantically aware sensor data acquisition strategy defines the working mode of the sensor under different regions and different semantic conditions, including parameters such as sampling frequency, data accuracy, and trigger conditions.

[0022] Specifically, according to the semantically-aware sensor data collection strategy and the real-time production rhythm of each area, the system activates the preset adaptive transmission protocol to schedule the transmission of the physical data collected by the sensor. During the implementation process, when the assembly area enters the high-precision assembly stage, the system automatically increases the transmission priority of the vibration sensor data in this area; when the injection molding area is in the mold replacement state, the data sampling rate and transmission frequency of the relevant sensors are reduced. The adaptive transmission protocol dynamically adjusts the data compression ratio, transmission priority and channel selection according to the semantic information, giving priority to the transmission of high-value data when the network is congested, and performing edge compression or delayed transmission on secondary data. In this way, the system obtains a transmission-optimized original data set, which effectively reduces the network bandwidth occupancy while ensuring the integrity of key quality data.

[0023] Specifically, the original data set that has been optimized for transmission is processed by three-dimensional association of region, equipment and data, and the data is associated with the corresponding regional semantic attributes, sensor characteristics and data collection background to obtain a structured quality data stream. In the implementation, three-dimensional tags are added to each data record: the region dimension records the semantic attributes of the data source area, such as "back-end assembly-refrigerator host-high precision requirements"; the equipment dimension records the characteristic parameters of the data collection equipment, such as "vibration sensor-high precision-5Hz sampling rate"; the data dimension records the context information of the collection environment, such as "normal production-second shift-36 product line". Through this three-dimensional association processing, the original physical quantity data is converted into a structured quality data stream with rich semantic information. Each data point carries complete context information, so that subsequent quality analysis can be carried out under the condition of fully understanding the data background. The structured quality data stream adopts a standardized data format to facilitate efficient transmission and processing between different system components.

[0024] 102. According to the structured quality data stream and preset quality inspection parameters, differential quality inspection load balancing distribution and multi-stop collaborative reasoning processing are performed on the logistics stop network within the factory to obtain semi-structured quality inspection result data containing defect type, location and severity information; In one embodiment of the present invention, the method of performing differential quality inspection load balancing allocation and multi-stop collaborative reasoning processing on the logistics stop point network within the factory according to the structured quality data stream and preset quality inspection parameters to obtain semi-structured quality inspection result data containing defect type, location and severity information includes: matching and analyzing the preset quality inspection parameters with the physical location and processing capacity of each logistics stop point in the logistics stop point network within the factory to construct a parameterized stop point resource distribution matrix; calculating and dynamically adjusting the quality inspection task allocation plan for each stop point according to the stop point resource distribution matrix and the structured quality data stream to achieve differential quality inspection load balancing allocation and obtain a load-balanced dynamic task allocation plan; executing quality inspection algorithms in parallel at each stop point according to the load-balanced dynamic task allocation plan and preset quality inspection parameters, and performing multi-stop collaborative reasoning for complex defects to obtain a decentralized preliminary quality inspection result set; and performing summary fusion and consistency verification processing on the decentralized preliminary quality inspection result set to form semi-structured quality inspection result data containing defect type, location and severity information.

[0025] Specifically, matching and analyzing the preset quality inspection parameters with the physical location and processing capacity of each logistics stop point in the logistics stop point network in the factory is the basis for constructing a parameterized stop point resource distribution matrix. The process first obtains the physical location information and hardware configuration information of all logistics stop points in the factory. For example, the A7 stop point on a home appliance production line is located in the northeast corner of the assembly area and is equipped with a high-performance edge computing unit to support image processing and deep learning reasoning; the B5 stop point is located in the center of the test area and is equipped with a medium-performance computing unit, which mainly supports numerical analysis. Then, the computing resources required for each type of quality inspection parameter are matched with the actual processing capacity of each stop point. For example, the surface defect detection parameter requires high image processing capability, which has a high matching degree with the A7 stop point; the electrical performance detection parameter has high requirements for numerical calculation, which has a high matching degree with the B5 stop point. Through this matching analysis, the system constructs a parameterized stop point resource distribution matrix, which contains the adaptability score of each stop point to various quality inspection tasks, providing a quantitative basis for subsequent task allocation.

[0026] According to the docking point resource distribution matrix and structured quality data flow, the system calculates and dynamically adjusts the quality inspection task allocation plan for each docking point to achieve balanced distribution of differential quality inspection loads. In actual operation, the system first preliminarily allocates the inspection task to the docking point closest to the data source based on the spatial distribution characteristics of the structured quality data flow. For example, vibration data from the assembly area is preferentially allocated to the A7 docking point for processing. Then the system considers the current load status of each docking point. For example, when the load rate of the A7 docking point has reached 80%, the system will reallocate some surface inspection tasks that should have been processed by A7 to the adjacent and compatible A8 docking point. The system will also dynamically adjust the task allocation according to the changes in the production line speed. For example, when a production line accelerates, the parallelism of task processing at the docking points along the line is increased accordingly. Through this multi-factor comprehensive calculation and real-time adjustment, the system obtains a load-balanced dynamic task allocation plan, which ensures that the load of each docking point is balanced, resources are fully utilized, and inspection tasks are processed in a timely manner.

[0027] According to the load-balanced dynamic task allocation scheme and the preset quality inspection parameters, the system executes the quality inspection algorithm in parallel at each docking point, and performs multi-dock collaborative reasoning for complex defects. In the specific implementation, each docking point receives the corresponding structured quality data stream fragment according to the allocation scheme, and applies the quality inspection algorithm suitable for its own processing capacity to process it. For example, the A7 docking point applies the deep learning algorithm to the received high-resolution image data to detect surface scratches, dents and other appearance defects; the B5 docking point applies the statistical analysis algorithm to the received electrical parameter data to detect performance anomalies. When encountering complex defects that are difficult to judge by a single docking point, the system starts the multi-dock collaborative reasoning mechanism. For example, when the A7 docking point detects that the appearance of a product is suspected to be abnormal but the confidence is not high, the system automatically extends the judgment task to the A8 and A9 docking points. The three docking points evaluate the same defect from different angles and determine the final conclusion through weighted voting. Through this parallel execution and collaborative reasoning, the system obtains a decentralized preliminary quality inspection result set, which contains the defect information independently judged and collaboratively judged by each docking point.

[0028] The process of aggregating, integrating and checking the consistency of the scattered preliminary quality inspection result sets to form semi-structured quality inspection result data involves multi-level data integration and verification. The system first classifies and integrates the preliminary quality inspection results from each stop point according to the product identification to build a complete defect profile for each product. Then, conflict detection is performed on the results of repeated inspection areas. For example, when the A7 and A8 stop points have inconsistent judgments on the defect type of the same area, the system performs weighted arbitration based on the historical accuracy and current judgment confidence of each stop point to determine the final judgment result. Then the system standardizes various defect information to uniformly represent the type code, spatial position coordinates, geometric size parameters and severity level of the defect. Finally, the system organizes the processed results into semi-structured quality inspection result data, which is stored in JSON format and includes the basic product information part, defect details part and detection meta-information part. It not only retains a strict data structure for program processing, but also has sufficient flexibility to meet the description requirements of different types of defects.

[0029] Furthermore, according to the dynamic task allocation scheme of load balancing and the preset quality inspection parameters, the quality inspection algorithm is executed in parallel at each stop, and multi-stop collaborative reasoning is performed for complex defects to obtain a decentralized preliminary quality inspection result set, including: according to the dynamic task allocation scheme of load balancing, the structured quality data stream is divided into multiple data subsets according to spatial distribution and data characteristics, and each data subset is distributed to the corresponding logistics stop to obtain a stop-level quality inspection data allocation scheme; according to the quality inspection data allocation scheme and the preset quality inspection parameters, feature extraction and defect pattern matching processing are performed on the allocated data subset at each stop, abnormal patterns are identified and the location and size of the defect are calculated to obtain a stop-level defect candidate set; for complex types of defects in the defect candidate set, a multi-stop collaborative judgment request is constructed, defect-related data is sent to adjacent stops, multi-source data fusion and arbitration processing are performed, and a collaboratively verified defect conclusion is obtained; the defect candidate set and defect conclusion of each stop are integrated into a quality inspection report in a unified format to obtain a decentralized preliminary quality inspection result set.

[0030] Specifically, dividing the structured quality data stream into multiple data subsets according to the load-balanced dynamic task allocation scheme is a prerequisite for parallel processing of quality inspection tasks. The system first preliminarily divides the data according to the principle of spatial distribution, and aggregates the data from the same or adjacent areas, such as grouping the data of the A1-A3 stations in the assembly area into one group and the data of the T1-T2 stations in the test area into another group. Then the system further subdivides the data according to the characteristics of the data, and classifies different types of data in the same area (such as image data, vibration data, and temperature data) according to their processing requirements. For example, on a refrigerator production line, the system divides the high-resolution image data of the A2 station in the assembly area into a subset, and merges the low-frequency vibration data of the same station and the similar data of the A1 station into another subset. After the division is completed, the system distributes each data subset to the corresponding logistics stop point according to the load-balanced dynamic task allocation scheme, such as allocating the image data subset to the AP3 stop point with GPU acceleration capability, and allocating the vibration data subset to the AP5 stop point that is good at spectrum analysis, thereby forming a stop point-level quality inspection data allocation scheme. The plan clearly defines the scope of data that each stop is responsible for processing, laying the data foundation for parallel quality inspection.

[0031] According to the inspection data allocation scheme at the docking point level and the preset inspection parameters, each docking point performs feature extraction and defect pattern matching processing on the assigned data subset. In the feature extraction stage, different algorithms are used for different types of data: for image data, the AP3 docking point uses convolutional neural network (CNN) to extract texture, edge and color features; for vibration data, the AP5 docking point applies fast Fourier transform (FFT) and wavelet transform to extract frequency domain features and time-frequency characteristics. In the defect pattern matching stage, each docking point compares the extracted features with the defect template defined in the preset inspection parameters. For example, the AP3 docking point uses template matching and support vector machine (SVM) algorithms to detect scratches, dents and color differences on the refrigerator panel; the AP5 docking point applies anomaly detection algorithms such as isolation forest to identify abnormal vibration patterns of the compressor. When a defect is detected, the docking point also calculates the precise position coordinates and geometric dimensions of the defect, such as the scratch's starting point coordinates (x1, y1), end point coordinates (x2, y2) and width w. Through these processes, each stop generates a list of identified defects, namely, the stop-level defect candidate set, which contains information such as defect type, location, size, and detection confidence.

[0032] For complex defects in the defect candidate set, the system constructs a multi-stop collaborative judgment request to achieve more accurate defect identification. Complex defects usually refer to defects that are difficult to determine at a single stop, such as defects with a detection confidence lower than a threshold (such as <0.75), or defects that require multi-angle data to confirm. When the AP3 stop detects a suspected but uncertain panel bubble defect, the system automatically constructs a collaborative judgment request, which contains defect location information, preliminary judgment results, and original image data. The system sends the request to adjacent stops that can provide supplementary evidence, such as the AP4 stop equipped with different light sources or angle cameras, and the AP7 stop that has historical defect statistics. After each collaborative stop receives the request, AP4 uses images under different lighting conditions for cross-validation and uses the ensemble learning method in deep learning to synthesize multi-angle evidence; AP7 uses the Bayesian inference model and combines historical defect statistical features to provide prior probability support. The judgment results of each stop are fused through a weighted voting algorithm, and the weights are dynamically adjusted based on the historical accuracy of each stop. Through this multi-source data fusion and arbitration processing, the system obtains a collaboratively verified defect conclusion, which includes the final defect type determination, comprehensive confidence level, and the determination basis of each collaborative point.

[0033] The final step in forming a decentralized preliminary quality inspection result set is to integrate the defect candidate set of each stop point and the defect conclusions verified by the collaboration into a quality inspection report in a unified format. The system first establishes a standardized data structure template, which contains three parts: product information area, defect list area and metadata area. Then the system collects the defect candidate sets generated by each stop point and merges different defect records belonging to the same product into the defect list of the corresponding product. For complex defects that have been verified by the collaboration, the system replaces its original defect candidate records with verified conclusions and attaches the process information of the collaborative verification. Then the system performs standardized conversion on the defect records, unifying the different representation methods that may be used by different stop points into the standard format specified by the system, such as converting the pixel coordinates used by AP3 into actual physical coordinates, and converting the frequency deviation values ​​used by AP5 into standardized severity levels (levels 1-5). Finally, the system adds metadata information to each quality inspection report, including the detection timestamp, the list of stop points involved in the detection, and the quality inspection parameter version used. Through these processes, the system generates a decentralized preliminary quality inspection result set, which, although distributed on each stop point, adopts a unified data structure and representation method.

[0034] 103. Based on the semi-structured quality inspection result data, perform topological path quality decline analysis and quality anomaly spatiotemporal clustering processing on the cargo locations and product quality status in the factory to obtain the corresponding quality knowledge graph; In one embodiment of the present invention, the topological path quality decline analysis and quality anomaly spatiotemporal clustering processing are performed on the cargo locations and product quality status in the factory according to the semi-structured quality inspection result data to obtain the corresponding quality knowledge graph, including: constructing a topological connection relationship model of the cargo locations in the factory according to the layout of the factory's logistics system and the semi-structured quality inspection result data; calculating the quality influence coefficient of each cargo location according to the topological connection relationship model and the semi-structured quality inspection result data to obtain a weighted cargo location quality influence network; using the cargo location quality influence network, calculating the quality change gradient of the product between each cargo location and detecting the path segment where the quality drops sharply, performing topological path quality decline analysis, and obtaining a set of quality critical paths; clustering the set of quality critical paths according to time and space dimensions to identify quality anomaly groups with similar spatiotemporal characteristics, and combining the cargo location quality influence network to construct the source node and diffusion path model of the quality problem to obtain a quality knowledge graph.

[0035] Specifically, building a topological connection relationship model of cargo locations in the factory based on the factory's logistics system layout and semi-structured quality inspection result data is the basis for understanding the quality transmission path. The construction process first collects the physical layout information of all cargo locations in the factory, including the spatial coordinates and relative position relationships of storage areas, production lines, assembly stations, test points, etc. For example, in a home appliance manufacturing plant, the system records the accurate location information of all cargo locations such as raw material warehouse area A, injection molding workshop area B, assembly line area C, and finished product warehouse area D. Then analyze the operating rules of the logistics system to determine the physical connection relationship between cargo locations, such as raw materials can only flow from cargo location A1 to cargo locations B1-B3, and products must flow from area B to area C through transfer station T1. Then the system digitizes the connection mode between each cargo location into a connection type identifier, such as conveyor belt connection is recorded as type 1, manual transfer is type 2, and automatic guided vehicle (AGV) transportation is type 3. Finally, the system calculates the time attribute of the connection and records the standard transmission time and actual average time from one cargo location to another. By integrating the above information, the system constructs a complete topological connection relationship model, which is essentially a directed weighted graph, in which nodes represent cargo locations, edges represent connection relationships, and edge weights contain multi-dimensional information such as distance, connection type, and time.

[0036] According to the topological connection relationship model and semi-structured quality inspection result data, the system calculates the quality impact coefficient of each cargo position and obtains the weighted cargo position quality impact network. The calculation process first extracts the product quality history record from the semi-structured quality inspection result data and tracks the quality status changes of each product at each quality inspection point. Then the system uses the statistical attribution analysis method to calculate the correlation between each cargo position and the product quality change. In specific implementation, the system applies multiple linear regression and random forest algorithms, takes cargo position characteristics (such as residence time, environmental conditions, and operation type) as independent variables, and takes quality change indicators as dependent variables to establish a regression model. For example, the analysis shows that the probability of surface scratch defects of products treated in the B2 cargo position increased by 5.2% in subsequent inspections, indicating that the cargo position has a negative impact on the surface quality. The system further calculates the quality impact coefficient matrix of each cargo position. Each element Eij of the matrix represents the degree of influence of cargo position i on quality indicator j. The value range is [-1,1]. Positive values ​​indicate positive influence and negative values ​​indicate negative influence. Through this calculation, the system obtains a weighted cargo location quality impact network, which adds the key attribute of quality impact coefficient to each node (cargo location) based on the original topological connection relationship, thus transforming the simple physical connection network into a quality impact network.

[0037] Using the shelf quality influence network, the system calculates the quality change gradient of the product between the shelves and detects the path segments where the quality drops sharply, and performs topological path quality decline analysis. The analysis process first reconstructs the complete movement trajectory of a large number of products from the semi-structured quality inspection result data, records the shelf sequence that each product passes through and the quality inspection results at each shelf. Then the system calculates the quality change gradient between adjacent shelves, that is, the change in the key quality index after the product moves from shelf A to shelf B is divided by the standard deviation to obtain the normalized quality change rate. In the gradient calculation, the system applies exponential smoothing technology to deal with data fluctuations, and uses the Z-score method to standardize the change amplitude of different quality indicators. Then the system identifies the path segments where the quality drops sharply, which are defined as shelf pairs where the quality change gradient exceeds the statistical threshold (-2σ). For example, the system detects that the assembly accuracy index drops by an average of 2.8 standard deviations when the product moves from shelf B3 to shelf C1, which is far beyond the normal fluctuation range. By applying graph algorithms such as Dijkstra's shortest path variant, the system searches for the path combination with the most serious quality degradation accumulation on the cargo network, and identifies the critical path with the most significant quality loss from raw materials to finished products. These paths together constitute the set of quality critical paths, which represent the logistics channels where quality problems are most likely to occur and accumulate in the factory.

[0038] The set of critical quality paths is clustered according to the time and space dimensions to identify the abnormal quality groups with similar time and space characteristics. Combined with the cargo location quality impact network, the source node and diffusion path model of the quality problem are constructed, and finally the quality knowledge graph is obtained. The clustering process first applies the DBSCAN (density-based spatial clustering) algorithm in the time dimension to identify the temporal clustering of quality problems, and classifies the abnormal quality paths that occur in similar time periods into the same category, and finds time patterns such as "the quality of products in area B generally declines after 3 pm every day". Then, the spectral clustering algorithm is applied in the spatial dimension to identify the spatial correlation of quality problems, classify the abnormal paths that affect similar cargo location groups into one category, and find spatial patterns such as "the B2, B3 and B5 cargo locations jointly affect the assembly quality of precision components". Then, the system combines the weight information of the cargo location quality impact network, applies causal inference techniques such as Bayesian network analysis, calculates the causal relationship of quality impact between each cargo location, and identifies the source node and diffusion path of the quality problem. For example, the system finds that the temperature anomaly of the B2 cargo location is the common starting point of multiple critical quality paths, and its influence spreads to multiple downstream cargo locations through a specific path. Finally, the system integrates all analysis results into a quality knowledge graph, which is a multi-level knowledge structure consisting of a cargo location node layer, a connection relationship layer, a quality impact layer, and a spatiotemporal pattern layer.

[0039] Furthermore, the method of using the cargo location quality influence network to calculate the quality change gradient of the product between each cargo location and detect the path segment where the quality drops sharply, and performing the topological path quality decline analysis to obtain the quality critical path set includes: reconstructing the movement trajectory of the product in the cargo location network according to the identification information and detection timestamp of the product in the semi-structured quality inspection result data, and obtaining the three-dimensional association sequence of product-cargo location-time; calculating the change amplitude of each key quality indicator of the product during the movement between adjacent cargo locations according to the three-dimensional association sequence and the cargo location quality influence network, constructing the quality change vector with the cargo location weight influence factor, and obtaining the quality change gradient set at the path segment level; applying the weight relationship in the cargo location quality influence network to the quality change gradient set, executing the weighted anomaly detection algorithm, and identifying the path segment where the quality change rate exceeds the weighted threshold; performing aggregation analysis on the path segment according to the connection relationship in the cargo location quality influence network to obtain the path combination, and calculating the cumulative quality risk score of each path combination to obtain the quality critical path set.

[0040] Specifically, reconstructing the movement trajectory of the product in the cargo network based on the product identification information and detection timestamp in the semi-structured quality inspection result data is the starting point for performing the topological path quality reduction analysis. The system first extracts the product unique identification code (such as the RFID number "FR20240307-A8642" of a refrigerator product), the detection location information and the detection timestamp from the semi-structured quality inspection result data. Then the system sorts all the detection records of the same product identification code in chronological order to form a time-series detection chain of the product. For example, the detection record of the product "FR20240307-A8642" shows that it passes through the raw material warehouse A5 cargo position (08:15), the injection molding area B2 cargo position (09:30), the assembly area C4 cargo position (10:45) and the test area D1 cargo position (11:30) in sequence. Then the system maps the detection location with the factory cargo topology network to determine all cargo positions that the product actually passes through, including detection points and non-detection points. When determining the location of non-detection points, the system uses the shortest path algorithm to analyze the intermediate storage locations that the product must pass through from one detection point to the next. By integrating this information, the system ultimately constructs a three-dimensional correlation sequence of product-storage location-time, which fully records the entire storage location record and corresponding time of each product from raw materials to finished products, forming a reproduction of the product's movement trajectory in physical space and time dimensions.

[0041] According to the three-dimensional correlation sequence of product-cargo location-time and the cargo location quality influence network, the system calculates the change range of each key quality indicator during the movement of the product between adjacent cargo locations, and constructs a quality change vector with cargo location weight influence factors. The calculation process first extracts the quality indicator value recorded at each inspection point from the semi-structured quality inspection result data. For example, the surface finish of the product "FR20240307-A8642" detected at the B2 cargo location is 92 points, and it drops to 85 points at the C4 cargo location. For the quality status of the intermediate cargo locations that cannot be directly obtained, the system uses linear interpolation and Markov prediction models for estimation. Then the system calculates the quality change between adjacent cargo locations, that is, the quality indicator value of the latter cargo location minus the value of the previous cargo location, to obtain the original change range. Then the system makes a weighted adjustment between the original change range and the influencing factors in the cargo location quality impact network. For example, the surface finish between cargo locations B2 and C4 drops by 7 points. Considering that the negative impact coefficient of cargo location B2 on surface quality is 0.8, the system calculates the adjusted change range to be 8.75 points, which more accurately reflects the impact of the path itself on quality. Through this calculation, the system constructs a quality change vector for each cargo location path. Each element in the vector corresponds to the change range of a quality indicator. Taking into account the quality impact characteristics of the cargo location itself, the quality change gradient set at the path segment level is obtained.

[0042] Applying the weighted relationship in the cargo location quality impact network to the quality change gradient set, the system executes a weighted anomaly detection algorithm to identify the path segments whose quality change rate exceeds the weighted threshold. The anomaly detection process first normalizes the quality change gradient and converts the changes in different quality indicators into Z scores, that is, the degree of deviation from the historical average change. Then the system applies an improved local anomaly factor (LOF) algorithm to identify the path segments with abnormal quality changes. The algorithm measures the degree of abnormality of each path segment by calculating the ratio of the local density of each path segment to the local density of its neighboring path segments. Unlike the traditional LOF algorithm, the improved version introduces the weighted relationship of the cargo location quality impact network as a weighting factor for density calculation, making anomaly detection more sensitive to cargo location paths with high quality impact. For example, the system detects that the surface finish Z score between cargo locations B2 and C4 is -3.2, which is far below the threshold of -2.0 set by the system, and the quality impact weight of this path is high, so it is marked as an abnormal path segment. Through this weighted anomaly detection, the system identifies a series of path segments whose quality change rate exceeds the weighted threshold, which represent the key links where product quality drops sharply.

[0043] The identified abnormal path segments are aggregated and analyzed according to their connection relationships in the cargo location quality impact network. The system obtains the path combination and calculates the cumulative quality risk score of each path combination, and finally forms a set of quality critical paths. The aggregation analysis first applies the depth-first search algorithm to find the complete path connecting adjacent abnormal path segments in the cargo location quality impact network. For example, the system finds that the abnormal path segments from B2 to C4 and the abnormal path segments from C4 to D1 are continuous in the network, so they are merged into a longer abnormal path from B2 to D1. Then the system applies community discovery algorithms such as the Louvain method to identify frequently co-occurring abnormal path segment combinations in the network to form a more complex path combination pattern. Then the system calculates the cumulative quality risk score of each path combination, which comprehensively considers the severity of the quality impact of each path segment in the path combination, the importance of the affected quality indicators, and the frequency of use of the path in production. For example, the B2-C4-D1 path combination affects multiple core quality indicators and is used frequently, and its cumulative risk score is calculated to be 8.7 (out of 10 points). Finally, the system sorts all path combinations according to the risk scores and selects path combinations with scores exceeding the threshold (such as 7.5 points) to form a quality critical path set.

[0044] 104. Based on the quality knowledge graph, multi-dimensional quality control limit self-learning adjustment and quality warning graded response processing are performed on the quality inspection parameters of different cargo locations and routes.

[0045] In one embodiment of the present invention, the self-learning adjustment of multi-dimensional quality control limits and quality warning graded response processing for quality inspection parameters of different cargo locations and paths according to the quality knowledge graph include: classifying the quality risk level of each cargo location in the factory according to the source node and diffusion path model in the quality knowledge graph to obtain a cargo location risk level distribution map; performing differentiated configuration on the quality inspection parameters in the logistics stop point network related to each cargo location according to the cargo location risk level distribution map to obtain a differentiated quality inspection parameter set; based on the differentiated quality inspection parameter set, combined with historical quality inspection data and actual quality results, using a preset parameter impact assessment algorithm to calculate the correlation strength between the quality inspection parameters and the quality indicators to obtain a quality inspection parameter impact weight matrix; according to the quality inspection parameter impact weight matrix, automatically adjusting the upper and lower control limits of the quality inspection parameters to obtain an adaptive quality inspection control limit set, and performing quality warning graded response processing according to the adaptive quality inspection control limit set and the diffusion path model of the quality knowledge graph.

[0046] Specifically, classifying the quality risk level of each cargo location in the factory according to the source node and diffusion path model in the quality knowledge graph is the basis for optimizing quality inspection parameters. The system first extracts the source node information from the quality knowledge graph and identifies the starting cargo location with a high probability of quality problems. For example, the B2 injection molding station on a home appliance production line is identified as the main source node of surface scratch defects, and its quality problem frequency is 3.7 times that of conventional stations. Then the system analyzes the diffusion path model to determine the propagation link and impact range of quality problems. For example, the quality problem of the B2 station mainly propagates along the B2→C4→D1 path, affecting multiple downstream processes. Then the system applies a variant of the PageRank algorithm to score the risk of each cargo location. The algorithm regards the quality knowledge graph as a directed weighted network, with cargo locations as nodes and quality impact relationships as edges, and calculates the risk importance score of each node. Finally, the system classifies all cargo locations into three levels according to the risk scores: cargo locations with scores in the top 20% are marked as high risk (red), cargo locations with scores between 20% and 50% are marked as medium risk (yellow), and the remaining cargo locations are marked as low risk (green). Through this classification, the system generates an intuitive cargo location risk level distribution map, which shows the quality risk level distribution of each area in the factory in the form of a heat map, clearly showing the spatial distribution of risk concentration areas and safe areas.

[0047] According to the distribution map of cargo location risk levels, the system performs differentiated configuration on the quality inspection parameters in the logistics stop network related to each cargo location to form a targeted quality control strategy. The configuration process first maps the cargo location risk level with its associated logistics stop, and determines the cargo location risk status that each stop is responsible for monitoring. For example, stop SP3 is mainly responsible for monitoring high-risk B2 cargo locations and medium-risk B3 cargo locations. Then the system adjusts the strictness of the quality inspection parameters according to the risk level. For stops related to high-risk cargo locations, the quality inspection sensitivity is improved, the allowable error range is reduced, and the detection frequency is increased; for medium-risk cargo locations, standard detection parameters are maintained; for low-risk cargo locations, the detection standards are appropriately relaxed and the detection frequency is reduced to save resources. The specific parameter adjustment follows the system's preset risk-parameter mapping rules. For example, for high-risk B2 stations, the pixel threshold for surface defect detection is adjusted from the standard 3 pixels to 2 pixels, and the sampling rate is increased from 1 in every 10 pieces to 1 in every 5 pieces. At the same time, the system considers the importance of different quality indicators and implements stricter parameter settings for key quality characteristics, such as electrical performance testing that affects product safety, and maintains high testing standards even in low-risk cargo locations. Through this comprehensive consideration, the system obtains a differentiated quality inspection parameter set, which customizes exclusive quality inspection standards for each stop, achieving the optimal configuration of quality inspection resources.

[0048] Based on the differential quality inspection parameter set, the system combines historical quality inspection data and actual quality results, and uses the preset parameter impact evaluation algorithm to calculate the correlation strength between quality inspection parameters and quality indicators to obtain the quality inspection parameter impact weight matrix. The evaluation process first extracts a large number of historical quality inspection records from the system database, including the inspection results and final product quality status when different quality inspection parameters are used. Then the system applies the gradient boosting decision tree (GBDT) algorithm to establish a nonlinear mapping relationship between quality inspection parameters and quality indicators. The algorithm captures complex parameter impact patterns by combining multiple weak learners. During the model training process, the system uses different quality inspection parameter combinations as feature inputs and the final quality score as the target output, and optimizes the decision tree set through continuous iteration. After the training is completed, the system uses the feature importance evaluation technology to extract the impact weight of each quality inspection parameter on the quality indicator. For example, the pixel threshold parameter of surface defect detection has an impact weight of 0.85 on the final appearance quality, and the impact weight on functional performance is only 0.12. By integrating the impact weights of all parameter-indicator pairs, the system constructs a comprehensive quality inspection parameter impact weight matrix, which accurately quantifies the potential impact of each quality inspection parameter adjustment on each quality indicator, providing a data basis for the precise tuning of quality inspection parameters.

[0049] According to the quality inspection parameter influence weight matrix, the system automatically adjusts the upper and lower control limits of the quality inspection parameters to obtain an adaptive quality inspection control limit set, and combines the diffusion path model of the quality knowledge graph to perform quality warning graded response processing. The parameter adjustment process first applies a multi-objective optimization algorithm such as NSGA-II (non-dominated sorting genetic algorithm II) to find the best combination of parameter control limits while meeting the quality goals. The algorithm continuously generates, evaluates and screens parameter solutions by simulating the evolutionary process, and finally converges to the optimal solution that balances quality detection rate and resource consumption. The optimization results form an adaptive quality inspection control limit set, which provides the system with a dynamically adjusted parameter range. For example, the pixel threshold for surface defect detection can float in the range of 1.8-2.5 pixels at different production stages. In the quality warning graded response processing link, the system constructs a four-level warning mechanism based on the diffusion path model of the quality knowledge graph: when an abnormality is detected in the source node, the system issues the highest level (red) warning and immediately shuts down for maintenance; when an abnormality occurs in the midstream node of the propagation link, a second-highest level (orange) warning is issued, the sampling frequency is increased and relevant personnel are notified; when an abnormality occurs in the terminal node, a medium-level (yellow) warning is issued, the abnormality is recorded and closely monitored; when a non-critical node experiences slight fluctuations, a low-level (blue) prompt is issued and it is included in the routine monitoring range.

[0050] In this embodiment, by performing regional semantic perception association on sensor combinations in different regions, and performing three-dimensional association processing on physical data through an adaptive transmission protocol, a structured quality data stream is obtained; based on the structured quality data stream and quality inspection parameters, differential quality inspection load balancing distribution and multi-stop collaborative reasoning are performed on the logistics stop network to obtain semi-structured quality inspection result data; based on the quality inspection result data, topological path quality reduction analysis and quality anomaly spatiotemporal clustering processing are performed on the cargo location and product quality status to obtain a quality knowledge graph; based on the quality knowledge graph, multi-dimensional quality control limit self-learning adjustment and quality warning graded response processing are performed on the quality inspection parameters of different cargo locations and paths. This method realizes the accurate tracing and rapid improvement of product quality problems, and improves the efficiency and accuracy of quality inspection.

[0051] The above describes the factory production quality detection method based on the Internet of Things in the embodiment of the present invention. The following describes the factory production quality detection system based on the Internet of Things in the embodiment of the present invention. Figure 2 In one embodiment of the present invention, a factory production quality inspection system based on the Internet of Things includes: The semantic association module 201 is used to perform regional semantic perception association on sensor combinations in different areas of the factory according to the regional division in the factory area modeling, and perform region-equipment-data three-dimensional association processing on the collected physical data through a preset adaptive transmission protocol to obtain a structured quality data stream; The collaborative inspection module 202 is used to perform differential quality inspection load balancing distribution and multi-stop collaborative reasoning processing on the logistics stop network in the factory according to the structured quality data stream and preset quality inspection parameters, and obtain semi-structured quality inspection result data containing defect type, location and severity information; The path analysis module 203 is used to perform topological path quality decline analysis and quality abnormality spatiotemporal clustering processing on the cargo locations and product quality status in the factory according to the semi-structured quality inspection result data, so as to obtain a corresponding quality knowledge graph; The parameter optimization module 204 is used to perform multi-dimensional quality control limit self-learning adjustment and quality warning graded response processing on the quality inspection parameters of different cargo locations and routes according to the quality knowledge graph.

[0052] In an embodiment of the present invention, the factory production quality detection system based on the Internet of Things runs the above-mentioned factory production quality detection method based on the Internet of Things, and the factory production quality detection system based on the Internet of Things performs regional semantic perception association on sensor combinations in different areas, and performs three-dimensional association processing on physical data through an adaptive transmission protocol to obtain a structured quality data stream; according to the structured quality data stream and quality inspection parameters, differential quality inspection load balancing distribution and multi-stop collaborative reasoning are performed on the logistics stop network to obtain semi-structured quality inspection result data; according to the quality inspection result data, topological path quality reduction analysis and quality abnormality spatiotemporal clustering processing are performed on the cargo location and product quality status to obtain a quality knowledge graph; according to the quality knowledge graph, multi-dimensional quality control limit self-learning adjustment and quality warning graded response processing are performed on the quality inspection parameters of different cargo locations and paths. This method realizes the accurate tracing and rapid improvement of product quality problems, and improves the efficiency and accuracy of quality inspection.

[0053] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described system, device, or unit can refer to the corresponding process in the aforementioned method embodiment and will not be repeated here.

[0054] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the whole or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.

[0055] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A factory production quality detection method based on the Internet of Things, characterized in that: The factory production quality detection method based on the Internet of Things includes: According to the regional division in the factory area modeling, the sensor combinations in different areas of the factory are associated with regional semantic perception, and the collected physical data is processed in three dimensions of area-equipment-data through the preset adaptive transmission protocol to obtain structured quality data stream; According to the structured quality data stream and preset quality inspection parameters, differential quality inspection load balancing allocation and multi-stop collaborative reasoning processing are performed on the logistics stop network in the factory to obtain semi-structured quality inspection result data containing defect type, location and severity information; According to the semi-structured quality inspection result data, topological path quality decline analysis and quality anomaly spatiotemporal clustering processing are performed on the cargo locations and product quality status in the factory to obtain the corresponding quality knowledge graph; According to the quality knowledge graph, multi-dimensional quality control limit self-learning adjustment and quality warning graded response processing are performed on the quality inspection parameters of different cargo locations and routes.

2. The factory production quality detection method based on the Internet of Things according to claim 1 is characterized in that: According to the regional division in the factory area modeling, the sensor combinations in different areas of the factory are associated with regional semantic perception, and the collected physical data are processed in three dimensions of area-equipment-data through a preset adaptive transmission protocol to obtain a structured quality data stream including: Semantically label the region according to the production function attributes of the factory region, extract the production stage, product type and quality requirement characteristics of the region, and construct the regional semantic feature vector; Mapping and associating the regional semantic feature vector with the existing sensor network in the region, assigning a regional semantic identity to each sensor, establishing a binding relationship between sensor data and regional semantics, and obtaining a semantically aware sensor data collection strategy; According to the semantically-aware sensor data collection strategy and the real-time production rhythm of each area, a preset adaptive transmission protocol is activated to schedule the transmission of the physical data collected by the sensor, so as to obtain a transmission-optimized original data set; The original data set is subjected to region-device-data three-dimensional association processing, and the data is associated and labeled with corresponding region semantic attributes, sensor characteristics, and data acquisition background to obtain a structured quality data stream.

3. The factory production quality detection method based on the Internet of Things according to claim 1 is characterized in that: According to the structured quality data stream and the preset quality inspection parameters, differential quality inspection load balancing distribution and multi-stop collaborative reasoning processing are performed on the logistics stop network in the factory to obtain semi-structured quality inspection result data containing defect type, location and severity information, including: Match the preset quality inspection parameters with the physical location and processing capacity of each logistics stop in the factory's logistics stop network to build a parameterized stop resource distribution matrix; According to the stop point resource distribution matrix and the structured quality data flow, the quality inspection task allocation plan of each stop point is calculated and dynamically adjusted to achieve differential quality inspection load balancing distribution and obtain a load-balanced dynamic task allocation plan; According to the load-balanced dynamic task allocation scheme and preset quality inspection parameters, the quality inspection algorithm is executed in parallel at each stop, and multi-stop collaborative reasoning is performed for complex defects to obtain a decentralized preliminary quality inspection result set; The scattered preliminary quality inspection result sets are aggregated, fused and checked for consistency to form semi-structured quality inspection result data containing defect type, location and severity information.

4. The factory production quality detection method based on the Internet of Things according to claim 3 is characterized in that: According to the load-balanced dynamic task allocation scheme and the preset quality inspection parameters, the quality inspection algorithm is executed in parallel at each stop, and multi-stop collaborative reasoning is performed for complex defects to obtain a decentralized preliminary quality inspection result set including: According to the load-balanced dynamic task allocation scheme, the structured quality data stream is divided into multiple data subsets according to spatial distribution and data characteristics, and each data subset is distributed to the corresponding logistics stop point to obtain a stop point level quality inspection data allocation scheme; According to the quality inspection data allocation scheme and the preset quality inspection parameters, feature extraction and defect pattern matching processing are performed on the allocated data subset at each stop point, abnormal patterns are identified and the position and size of defects are calculated to obtain a defect candidate set at the stop point level; For complex defects in the defect candidate set, a multi-stop collaborative determination request is constructed, defect-related data is sent to adjacent stops, multi-source data fusion and arbitration processing are performed, and a collaboratively verified defect conclusion is obtained; The defect candidate sets and defect conclusions of each stop point are integrated into a quality inspection report in a unified format to obtain a decentralized preliminary quality inspection result set.

5. The factory production quality detection method based on the Internet of Things according to claim 1 is characterized in that: According to the semi-structured quality inspection result data, the topological path quality decline analysis and quality abnormality spatiotemporal clustering processing are performed on the cargo locations and product quality status in the factory to obtain the corresponding quality knowledge graph including: According to the layout of the factory's logistics system and semi-structured quality inspection result data, a topological connection relationship model of the cargo locations in the factory is constructed; According to the topological connection relationship model and semi-structured quality inspection result data, the quality impact coefficient of each cargo location is calculated to obtain the weighted cargo location quality impact network; By using the cargo location quality influence network, the quality change gradient of the product between the cargo locations is calculated and the path segments where the quality drops sharply are detected, and a topological path quality decrease analysis is performed to obtain a set of quality critical paths; The quality critical path set is clustered according to the time and space dimensions to identify quality anomaly groups with similar time and space characteristics. In combination with the cargo location quality impact network, the source node and diffusion path model of the quality problem are constructed to obtain a quality knowledge graph.

6. The factory production quality detection method based on the Internet of Things according to claim 5 is characterized in that: The quality impact network of the cargo location is used to calculate the quality change gradient of the product between the cargo locations and detect the path segments where the quality drops sharply, and the topological path quality decrease analysis is performed to obtain the quality critical path set including: According to the product identification information and detection timestamp in the semi-structured quality inspection result data, the movement trajectory of the product in the storage location network is reconstructed to obtain the three-dimensional correlation sequence of product-storage location-time; According to the three-dimensional association sequence and the cargo location quality influence network, the change range of each key quality indicator of the product during the movement between adjacent cargo locations is calculated, a quality change vector with a cargo location weight influence factor is constructed, and a quality change gradient set at the path segment level is obtained; Applying the weight relationship in the cargo location quality impact network to the quality change gradient set, executing a weighted anomaly detection algorithm, and identifying a path segment whose quality change rate exceeds a weighted threshold; The path segments are aggregated and analyzed according to their connection relationships in the cargo location quality influence network to obtain path combinations, and the cumulative quality risk scores of each path combination are calculated to obtain a quality critical path set.

7. The factory production quality detection method based on the Internet of Things according to claim 1 is characterized in that: The performing of multi-dimensional quality control limit self-learning adjustment and quality warning graded response processing on quality inspection parameters of different cargo locations and paths according to the quality knowledge graph includes: According to the source nodes and diffusion path model in the quality knowledge graph, the quality risk level of each cargo location in the factory is classified to obtain the cargo location risk level distribution map; According to the cargo location risk level distribution map, performing differentiated configuration on quality inspection parameters in the logistics stop point network related to each cargo location to obtain a differentiated quality inspection parameter set; Based on the differential quality inspection parameter set, combined with historical quality inspection data and actual quality results, a preset parameter impact assessment algorithm is used to calculate the correlation strength between the quality inspection parameters and the quality indicators to obtain a quality inspection parameter impact weight matrix; According to the quality inspection parameter influence weight matrix, the upper and lower control limits of the quality inspection parameters are automatically adjusted to obtain an adaptive quality inspection control limit set, and quality warning graded response processing is performed according to the adaptive quality inspection control limit set and the diffusion path model of the quality knowledge graph.

8. A factory production quality inspection system based on the Internet of Things, characterized in that: The factory production quality inspection system based on the Internet of Things includes: The semantic association module is used to perform regional semantic perception association on sensor combinations in different areas of the factory according to the regional division in the factory area modeling, and perform region-equipment-data three-dimensional association processing on the collected physical data through a preset adaptive transmission protocol to obtain a structured quality data stream; A collaborative inspection module, which is used to perform differential quality inspection load balancing distribution and multi-stop collaborative reasoning processing on the logistics stop network in the factory according to the structured quality data stream and preset quality inspection parameters, and obtain semi-structured quality inspection result data containing defect type, location and severity information; A path analysis module is used to perform topological path quality decline analysis and quality anomaly spatiotemporal clustering processing on the cargo locations and product quality status in the factory according to the semi-structured quality inspection result data, so as to obtain a corresponding quality knowledge graph; The parameter optimization module is used to perform multi-dimensional quality control limit self-learning adjustment and quality warning graded response processing on the quality inspection parameters of different cargo locations and routes according to the quality knowledge graph.

Images