Production data hierarchical processing and quality tracing system

Abstract

The invention relates to the technical field of production quality control, and discloses a production data hierarchical processing and quality tracing system. The system comprises a production data acquisition module, a data hierarchical processing module, a quality fluctuation index module, a multi-dimensional quality difference analysis module and an abnormal source positioning module. The production data acquisition module acquires a real-time parameter data sequence of each production link; the data layering processing module is used for layering the data into a plurality of data hierarchies; the quality fluctuation index module obtains a quality fluctuation index based on the fluctuation condition of the data hierarchy; the multi-dimensional quality difference analysis module analyzes the multi-dimensional quality difference based on the index; and the abnormal source positioning module positions a production abnormal source based on the multi-dimensional quality difference. According to the system, through hierarchical data processing, quality fluctuation quantification and multi-dimensional difference analysis, accurate positioning of a production abnormal source is realized, and systematicness and accuracy of production data processing and quality tracing are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of production quality control, in particular to a production data hierarchical processing and quality traceability system. BACKGROUND

[0002] With the continuous improvement of industrial automation level, the amount of data generated in the production process increases exponentially, covering equipment operation parameters, material ratio data, environmental temperature and humidity, process execution time, etc. These data are distributed in multiple production links such as procurement, processing, assembly, detection, etc., and there are complex relationships between them. However, the current production data management mode generally lacks effective hierarchical division mechanism, and different types of data are often randomly stacked, resulting in chaotic data structure and difficulty in quickly extracting key information. In actual production, this data management method will cause a series of problems. For example, when a batch of products has quality deviation, the manager needs to filter the relevant information from the massive data one by one, and due to the lack of hierarchical classification, a large amount of irrelevant data will interfere with the analysis process, making the perception of quality fluctuation lag. At the same time, the existing system only stays in a single dimension in data analysis, such as only focusing on equipment parameter changes or only tracking material characteristics, and cannot establish a correlation model between multiple factors, leading to one-sided cognition of quality differences. The traditional quality traceability process relies too much on manual recording and experience judgment. When there is a quality anomaly, data needs to be retrieved across departments and coordination meetings need to be held, which not only makes the process cumbersome, but also easily leads to misjudgment of the source of the anomaly due to information transmission bias. This inefficient traceability method makes it difficult to correct potential problems in the production process in a timely manner, which may cause batch quality problems and affect the smooth progress of the production plan. SUMMARY

[0003] The purpose of the present application is to provide a production data hierarchical processing and quality traceability system to solve the problems raised in the background.

[0004] To achieve the above purpose, the present application provides a production data hierarchical processing and quality traceability system, which comprises: a production data acquisition module, a data hierarchical processing module, a quality fluctuation index module, a multi-dimensional quality difference analysis module, and an anomaly source positioning module; The production data acquisition module is used to acquire real-time parameter data sequences of each production link; The data hierarchical processing module is used to hierarchically process the real-time parameter data sequences into multiple data levels; The quality fluctuation index module is used to obtain a quality fluctuation index based on the data fluctuation in the multiple data levels; The multi-dimensional quality difference analysis module is configured to analyze the multi-dimensional quality difference based on the quality fluctuation index. The abnormal source positioning module is configured to locate the production abnormal source based on the multi-dimensional quality difference.

[0005] Preferably, the production data acquisition module comprises a parameter data acquisition unit configured to acquire temperature sequences, pressure sequences and speed sequences of each production link at each acquisition time of each production cycle. The reference data acquisition unit is configured to acquire historical parameter data sequences of each production link at each production cycle.

[0006] Preferably, the data hierarchical processing module comprises an original data layer processing unit configured to store the real-time parameter data sequences. A feature extraction layer processing unit is configured to extract feature parameter sequences from the original data layer processing unit. An aggregation analysis layer processing unit is configured to aggregate the feature parameter sequences into high-order data sequences; the output of the feature extraction layer processing unit is connected to the input of the aggregation analysis layer processing unit.

[0007] Preferably, the quality fluctuation index module comprises a fluctuation point identification unit configured to identify each fluctuation point in the feature parameter sequences. A fluctuation difference calculation unit is configured to acquire a forward difference index based on the difference between the fluctuation points. A cluster index calculation unit is configured to acquire a cluster index based on the difference of the feature parameter sequences. A mean difference calculation unit is configured to acquire a before-after mean difference based on the mean difference of each local feature parameter sequence. A sudden change index generation unit is configured to acquire a quality fluctuation index of each fluctuation point based on the forward difference index, the cluster index and the before-after mean difference.

[0008] Preferably, the quality fluctuation index module further comprises an average quality fluctuation index generation unit configured to acquire an average quality fluctuation index of each fluctuation acquisition time of each production cycle based on the average of the quality fluctuation indexes of the fluctuation points. The input of the average quality fluctuation index generation unit receives the output of the sudden change index generation unit.

[0009] Preferably, the quality fluctuation index module further comprises a production abnormal index generation unit configured to acquire a production abnormal index of each fluctuation acquisition time of each production cycle based on the distance between local feature parameter sequences and the average quality fluctuation index. The input of the production abnormal index generation unit receives the output of the average quality fluctuation index generation unit.

[0010] Preferably, the multi-dimensional quality difference analysis module comprises: a quality sensitivity coefficient generation unit configured to obtain quality sensitivity coefficients of each production link in each production cycle based on the production anomaly index; a parameter deviation coefficient generation unit configured to obtain parameter deviation coefficients based on deviation conditions of the historical parameter data sequence of the reference data acquisition unit; a quality control parameter generation unit configured to obtain quality control parameters based on the quality sensitivity coefficients and the parameter deviation coefficients; The input of the quality control parameter generation unit receives the outputs of the quality sensitivity coefficient generation unit and the parameter deviation coefficient generation unit.

[0011] Preferably, the multi-dimensional quality difference analysis module further comprises: a multi-dimensional difference analysis unit configured to perform multi-dimensional quality difference analysis based on the quality control parameters, wherein the multi-dimensional quality difference analysis comprises time-domain cumulative deviation analysis, frequency-domain energy offset analysis, and sequence similarity evaluation; a difference coefficient matrix generation unit configured to generate a difference coefficient matrix based on the time-domain cumulative deviation analysis, the frequency-domain energy offset analysis, and the sequence similarity evaluation; The input of the difference coefficient matrix generation unit receives the output of the multi-dimensional difference analysis unit.

[0012] Preferably, the abnormal source positioning module comprises: a production topology modeling unit configured to construct a production topology network based on position information of production links; an abnormal propagation simulation unit configured to perform abnormal propagation simulation based on the difference coefficient matrix and the production topology network; a probability distribution generation unit configured to generate an abnormal probability distribution map based on the abnormal propagation simulation; a physical region positioning unit configured to position a production abnormal physical region based on the abnormal probability distribution map; The input of the physical region positioning unit receives the output of the probability distribution generation unit.

[0013] Preferably, the abnormal source positioning module further comprises: a verification strategy generation unit configured to configure a production verification strategy based on the abnormal probability distribution map, wherein the production verification strategy comprises high-frequency monitoring mode enabling and node disturbance test execution; The input of the verification strategy generation unit receives the output of the physical region positioning unit.

[0014] Compared with the prior art, the present application has the following advantages: By setting the production data acquisition module, real-time parameter data sequences of each production link can be comprehensively collected, providing complete basic information for subsequent data processing and analysis. The data hierarchical processing module processes the real-time parameter data sequences in layers, dividing multiple data levels. This hierarchical approach helps to distinguish data of different importance and correlation attributes, making the data processing more targeted and avoiding information interference and lack of emphasis when all data are mixed together. The quality fluctuation index module obtains the quality fluctuation index based on the data fluctuation in multiple data levels, which can integrate scattered quality fluctuation information and form a quantifiable index to intuitively reflect the quality fluctuation state in the production process. The formation of this index enables production managers to more clearly understand the overall trend of quality changes and break free from the limitations of relying on subjective judgment or scattered data for evaluation. The multi-dimensional quality difference analysis module performs multi-dimensional quality difference analysis based on the quality fluctuation index, breaking through the limitations of traditional single-dimensional analysis and deeply analyzing the performance and characteristics of quality differences from multiple angles. Through this multi-dimensional analysis, the quality correlation between different production links, different time periods, and different parameters can be fully revealed, enabling production managers to have a more comprehensive and in-depth understanding of the quality status and discover quality problems that are easily overlooked in single-dimensional analysis. The abnormal source positioning module locates the abnormal source based on multi-dimensional quality difference, which can accurately lock the specific link or factor causing quality abnormalities. This accurate positioning avoids blind investigation in traditional traceability process, reduces the waste of investigation time and energy, and enables quality problems in the production process to be quickly solved, which helps to timely adjust production parameters or improve production processes, maintain the stability and continuity of production. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 The timing diagram of the production data hierarchical processing and quality traceability system described in the present application; Figure 2 The flowchart of the data hierarchical processing module; Figure 3 The flowchart of the quality fluctuation index calculation; Figure 4 The flowchart of the quality control parameter generation; Figure 5 The flowchart of the abnormal source positioning module. DETAILED DESCRIPTION

[0016] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of the present application, but not all the embodiments of the present application. Based on the embodiments of the present application, all the other embodiments obtained by a person of ordinary skill in the art without creative work are within the protection scope of the present application.

[0017] With reference to Figure 1 The present application provides a production data hierarchical processing and quality traceability system, which comprises: through hierarchical processing of production data and combining multi-dimensional analysis technology, accurate positioning of the source of production abnormalities is realized. The system comprises a production data acquisition module, a data hierarchical processing module, a quality fluctuation index module, a multi-dimensional quality difference analysis module and an abnormal source positioning module.

[0018] The production data acquisition module is responsible for acquiring real-time parameter data sequences of each production link, the data hierarchical processing module processes the real-time parameter data sequences into multiple data levels, the quality fluctuation index module calculates the quality fluctuation index based on the data fluctuation in the data levels, the multi-dimensional quality difference analysis module analyzes the multi-dimensional quality difference through the quality fluctuation index, and the abnormal source positioning module locates the source of production abnormalities based on the multi-dimensional quality difference.

[0019] Embodiment 1: With reference to Figure 2 The production data acquisition module captures the time sequence parameters of each production link in real time through a parameter data acquisition unit. The unit is deployed at the key nodes of the production line, the temperature sequence is collected by thermocouple sensors distributed on the surface of the equipment, the sampling frequency is synchronized with the production rhythm, and a temperature data stream arranged according to millisecond timestamps is formed in each production cycle. The pressure sequence is obtained through an embedded piezoelectric sensor, the sensor is integrated in the contact interface between the hydraulic actuator and the material, and the pressure change curve is recorded in real time. The speed sequence is collected by a rotary encoder, the encoder is installed at the end of the transmission shaft, and the angular displacement is converted into linear speed values. All sensor data is transmitted to the edge computing node through the industrial bus and stored as a structured data set according to the production cycle number, each data point contains a timestamp, a parameter type, a numerical value and an identification code of the production link. The reference data acquisition unit is connected to a historical database, the database establishes a three-dimensional index structure according to the production link, equipment number and product batch, automatically matches the historical records of the same production line configuration according to the current production cycle during retrieval, and extracts the complete sequences of temperature, pressure and speed of the same type of product in the past three months as a reference set.

[0020] The data hierarchical processing module constructs a three-level data processing pipeline. The raw data layer processing unit adopts a time series database architecture. After receiving real-time parameter streams from the acquisition module, it establishes independent data partitions according to production links. The data in each partition is stored in a ring buffer format, and the buffer capacity covers the maximum data volume of a single production cycle. A globally unique time series identifier is automatically attached when writing. The feature extraction layer processing unit processes raw data through a sliding window mechanism. The window width is dynamically adjusted according to the production rhythm: a fixed 10-second window is used for continuous process sections, and an event-triggered window is used for discrete processes. Multi-dimensional feature calculation is performed within the window, including but not limited to statistical features (window mean, standard deviation, skewness), morphological features (zero-crossing rate, waveform factor), and time-domain features (autocorrelation coefficient). The calculation results form a feature vector sequence, and the vector elements include feature type encoding, time interval marker, and feature value. The aggregation analysis layer processing unit receives the feature vector stream and performs spatial and temporal aggregation: in the spatial dimension, group by process section, and fuse the feature vectors of multiple devices in the same process section; in the time dimension, use the piecewise linear representation method to compress the long sequence into a simplified sequence composed of key turning points. The high-order data sequence output by this unit is stored in a tree index structure, with the root node being the production cycle identifier, the child nodes being layered by process section, and the leaf nodes storing the aggregated feature summary data. The feature extraction layer and the aggregation analysis layer are connected through a message queue to realize asynchronous decoupled data transmission. The queue sets a priority strategy to ensure that critical process data is processed first.

[0021] In specific implementation, the temperature sequence processing adopts adaptive filtering technology to eliminate environmental interference, the pressure sequence is separated into static load and dynamic fluctuation components through wavelet transform, and the speed sequence is smoothed through Kalman filtering. The historical parameter retrieval adopts a similarity matching algorithm, calculates the matching degree of the current production conditions and historical records through dynamic time warping, and selects the top five cycle data with the highest similarity as the reference set. The original data storage adopts columnar compression storage, and independent storage columns are established for temperature, pressure, and speed data to improve the efficiency of time series range queries. In the feature extraction stage, different processing strategies are configured for different parameter types: temperature data focuses on extracting rising / descending slope features, pressure data focuses on peak duration features, and speed data mainly focuses on cycle stability features. High-order data aggregation uses a streaming processing framework, with both time-triggered and quantity-triggered mechanisms. When the accumulated data reaches a certain threshold or a fixed time interval, the aggregation operation is automatically triggered. The process section grouping is based on a mapping table configured according to the physical layout of the production line, which can be updated online to adapt to production line reorganization. Resource isolation strategies are implemented throughout the hierarchical processing process to allocate independent computing resources to critical processes, avoiding low-priority task blocking of real-time processing links. When data is transmitted between levels, a complete integrity check code is attached to ensure error-free data transmission using cyclic redundancy check.

[0022] Embodiment 2: refer to Figure 3 The quality fluctuation index module quantifies the abnormal fluctuation of the feature parameter sequence through a multi-stage calculation process. After receiving the feature vector sequence from the feature extraction layer processing unit, the fluctuation point recognition unit first performs data smoothing preprocessing, using a filtering method based on local weighted regression to eliminate high-frequency noise interference. The recognition process uses an adaptive window scanning mechanism, and the window width is dynamically adjusted according to the sampling density of the feature sequence: a smaller window is set for high sampling rate data to capture transient fluctuations, and a larger window range is set for low sampling rate data. Local extreme points are calculated within each window, and valid fluctuation points are identified by setting double threshold judgment conditions: the data point must meet both the absolute change threshold relative to the previous data and exceed the relative threshold of the change rate of adjacent data points. The identified fluctuation points are marked with a timestamp, a fluctuation direction identifier, and a fluctuation amplitude value, forming a fluctuation event set.

[0023] The fluctuation difference calculation unit processes the fluctuation event set by establishing a forward difference analysis model. This model traverses the fluctuation points in chronological order, calculates the change gradient between adjacent fluctuation points, and focuses on the transition characteristics of the rising edge and the falling edge. For a sequence of continuous rising or continuous falling fluctuation points, the cumulative change acceleration is calculated; for fluctuation points with alternating directions, the mutation strength of the turning point is analyzed. The forward difference index is generated by quantifying the morphological difference between fluctuation points, containing three-dimensional indicators: adjacent fluctuation amplitude ratio, fluctuation duration ratio, and fluctuation shape similarity. The fluctuation shape similarity is calculated by the dynamic time warping algorithm to calculate the similarity of adjacent fluctuation curves, avoiding misjudgment caused by time axis stretching. The calculation result forms a weighted forward difference index vector, and the weight allocation is based on the importance of the fluctuation point in the time sequence, with recent fluctuations being given higher weights.

[0024] The cluster index calculation unit processes the overall distribution characteristics of the feature parameter sequence in parallel. This unit constructs a statistical distribution model of the sequence, first calculating global statistics including overall mean, standard deviation, and skewness coefficient. Then, using a sliding binning strategy, the sequence is divided into several intervals according to the percentile, and the data point density in each interval is calculated. The cluster index is obtained by measuring the density matching degree of a single data point with the interval it belongs to, and the specific calculation includes two complementary indicators: local density deviation and global position offset. The local density deviation reflects the degree of agreement between the data point and the adjacent point distribution, and the global position offset measures the relative position of the point value in the overall distribution. Finally, the cluster index is expressed in the form of the harmonic mean of the density deviation and the position offset, and the numerical range is normalized to the standard interval. The calculation process uses a streaming update mechanism, and when a new data point arrives, only the relevant statistics need to be updated incrementally, avoiding full sequence recalculation.

[0025] The mean difference calculation unit focuses on the mean change characteristics of the local data segment. A symmetrical window before and after is used as the analysis unit, and the time interval is expanded in the forward and backward directions respectively with the current fluctuation point as the center. The forward window covers the steady-state data before the fluctuation occurs, and the backward window contains the evolution data after the fluctuation occurs. The window size is adaptively determined according to the fluctuation duration, and the minimum is not less than three sampling points. After calculating the arithmetic mean of the front and rear windows, the standardized mean difference algorithm is used to eliminate the dimension influence, and a data stability correction factor is introduced: when the standard deviation in the window is large, the confidence weight of the mean difference is appropriately reduced. For the case that the fluctuation point is located at the boundary of the sequence, the asymmetric window processing mode is started, and the virtual data points are constructed by the mirror filling method to ensure the calculation integrity.

[0026] The sudden change index generation unit integrates the output results of the first three units for fusion calculation. The unit establishes a three-dimensional input space: the forward difference index represents the relative change strength between fluctuation points, the clustering index reflects the coordination between fluctuation points and the overall sequence, and the mean difference before and after indicates the offset degree of local data. The fusion process adopts a variable weight strategy, and the weight coefficient is dynamically adjusted according to the production stage: the clustering index weight is increased in the steady-state production stage, and the forward difference index is focused on in the process switching stage. After the preliminary fluctuation index is generated by fusion calculation, a time decay factor is introduced for time series smoothing processing, so that recent fluctuations obtain higher index values. The final output quality fluctuation index is attached with a confidence label, and the confidence is determined by the original data quality involved in the calculation, which automatically reduces the confidence level when there are missing or abnormal input data. The entire calculation process implements an abnormal isolation mechanism, and automatically switches to the backup algorithm when any sub-unit calculation fails, avoiding single-point failure that interrupts index generation.

[0027] In the specific implementation layer, the fluctuation point recognition adopts a multi-thread parallel processing architecture, and independent calculation threads are allocated for different production links. The forward difference analysis introduces a direction-sensitive coefficient, and implements an asymmetric processing strategy for rising and falling fluctuations. The clustering index calculation sets an adaptive number of bins, and automatically increases the number of bins to improve the resolution accuracy when the data complexity is high. The mean difference calculation configures differential processing parameters for different types of parameters: temperature data focus on slow drift, pressure data focus on instantaneous mutation, and speed data value periodic deviation. The sudden change index fusion stage implements hierarchical weighting, and gives global amplification coefficients to the fluctuation index of key process links. All intermediate calculation results are marked with time validity, and data that is not updated in time automatically triggers the recalculation process. The index generation process implements full-link tracking, and each fluctuation index can be traced back to the original fluctuation event and the feature data points involved in the calculation. The calculation resource allocation adopts priority queue management, and the data of core processes that affect the final product quality are processed preferentially. The data caching mechanism ensures the index generation frequency under system peak load, avoiding data loss or calculation delay.

[0028] Example 3: After generating the quality fluctuation index of each fluctuation point, the quality fluctuation index module implements an index aggregation process through an average quality fluctuation index generation unit. This unit receives a stream of quality fluctuation indices from the sudden change index generation unit, which are arranged in chronological order and attached with timestamps, production link identifiers, and confidence markers. The aggregation process uses a dynamic sliding window mechanism, with the window time span adjusted synchronously with the production rhythm: a larger time window is set for stable production stages to smooth random fluctuations, and the window range is reduced during process conversion stages to retain detailed features. Weighted average calculation is performed within the window, with weight distribution following a dual criterion: the time decay criterion gives higher weight to recent indices, and the confidence criterion gives stronger influence to high-confidence data. The specific calculation expression is:

[0029] where: represents the average quality fluctuation index at time t, N is the number of indices within the window, is the kth quality fluctuation index original value, is the corresponding confidence factor (value range 0.1-1.0), is the time decay weight (calculated according to the exponential decay function , t K is the index generation time, is the decay coefficient). The calculation result is attached with time interval markers and the number of original indices involved in the calculation, forming an average index sequence with time dimension.

[0030] The production anomaly index generation unit receives the average quality fluctuation index sequence while processing the feature parameter sequence from the feature extraction layer processing unit in parallel. This unit first constructs a feature distance calculation model, using differentiated distance measurement strategies for different feature types: standardized Euclidean distance for statistical features, dynamic time warping distance for morphological features, and spectral correlation distance for frequency domain features. Distance calculation takes the current production cycle feature sequence as the reference, simultaneously calculating three types of comparison distances: horizontal distance (compared with historical reference sequence of the same link), vertical distance (compared with current sequence of the upstream link), and baseline distance (compared with ideal process parameter template). The distance calculation result is normalized to a similarity score in the [0, 1] interval.

[0031] The anomaly index synthesis stage establishes a three-dimensional fusion space: the first dimension inputs the average quality fluctuation index , the second dimension inputs the feature similarity comprehensive score (obtained by weighted fusion of three types of distance scores), and the third dimension introduces environmental factors (including equipment running time, environmental temperature and humidity, and other working condition parameters). The fusion process uses a non-linear mapping function:

[0032] wherein: is the production anomaly index, a is the fluctuation index weight coefficient (dynamically adjusted between 0.4-0.7 according to the production stage), b is the index amplification factor (default value is 1.2), is the minimum constant to prevent division by zero error. Environmental factor Through the fuzzy logic system calculation, the continuous working condition parameters are converted into the correction coefficient in the interval [0.8, 1.2]. The calculation result After Gaussian filtering smoothing processing, the additional time stamp and production link code are stored in the distributed index database.

[0033] In specific implementation, the average index calculation window implements intelligent expansion control: when the window index variance exceeds the threshold value, the window size is automatically reduced, and when continuous stable state is detected, the window range is gradually expanded. Confidence factor The generation considers three dimensions of data acquisition quality, feature extraction integrity and calculation process reliability, and comprehensively determines through decision tree rules. The feature distance calculation uses a pre-computation acceleration strategy, and the historical reference sequence is constructed in advance. The feature index tree is used to improve the efficiency through approximate nearest neighbor search online. The environmental factor model includes a device operating state mapping table, which is dynamically updated according to the device maintenance record, and automatically improves the environmental factor sensitivity of the device close to the maintenance cycle.

[0034] The parameters in the abnormal index synthesis formula implement an online learning mechanism: the weight coefficient a is dynamically optimized through a reinforcement learning framework, and the parameter value is adjusted based on the previous day anomaly verification result every 24 hours. The calculation process implements multiple checks: when and there is a significant deviation, the manual review process is started; when the input data has a time synchronization problem, the data alignment and recalculation are triggered. The resource allocation uses a dynamic scheduling based on the abnormal index prediction, and the production link with obvious index rising trend is preferentially allocated computing resources. The data storage uses a hybrid strategy of columnar compression and time partitioning, which supports fast retrieval of historical abnormal index sequence by time range. All calculation components are deployed as microservices architecture, and each production link independently runs a calculation instance, and realizes data exchange through the message bus. The service instance implements health monitoring, and automatically switches to the standby calculation node when abnormal, and the calculation state is written into the blockchain storage system in real time to ensure process traceability.

[0035] Embodiment 4: see Figure 4, the multi-dimensional quality difference analysis module is implemented, and a specific process is illustrated by taking an injection molding production line as an example. The production line includes three core links of plasticizing, injection, and pressure maintaining. The quality sensitivity coefficient generation unit receives the production anomaly index sequence from example 3. For the current production cycle of the plasticizing link, the anomaly index sequence shows that a peak value 0.85 appears in the time interval [09:30:15, 09:30:45]. The unit starts the sensitivity conversion program: first, the anomaly index is normalized in the time dimension, and 0.85 is mapped to the standard range [0, 1] to obtain 0.92; then, a linear conversion formula is applied, and the process weight factor 0.7 of the plasticizing link (the factor is stored in the process knowledge base) is combined to generate the final quality sensitivity coefficient 0.644. Similarly, the peak value 0.78 of the injection link is processed to obtain the coefficient 0.68, and the peak value 0.62 of the pressure maintaining link is processed to obtain the coefficient 0.55.

[0036] The parameter deviation coefficient generation unit operates synchronously and extracts historical parameter data from the reference data acquisition unit. Taking the temperature parameter of the plasticizing link as an example, the temperature sequence of the last 30 normal production cycles is retrieved, and the statistical distribution of each sampling point is calculated. The actual measured value of the temperature at 09:30:20 in the current cycle is 245°C, which is compared with the historical mean value of 238°C, and the standard deviation is ±5°C. The deviation calculation is processed in sections: within the interval [238-5, 238+5], it is considered as normal deviation, and the current value 245°C exceeds the upper limit by 3°C. According to the preset deviation-coefficient mapping table, each 1°C of the exceeding part corresponds to an increment of the deviation coefficient of 0.15, so the temperature deviation coefficient of this point is 0.45. Similarly, the pressure parameter deviation coefficient is calculated as 0.32, and the speed parameter deviation coefficient is calculated as 0.18. The final parameter deviation coefficient is the weighted average of the three, with the weight distribution being temperature 0.6, pressure 0.3, and speed 0.1, and the comprehensive parameter deviation coefficient of the plasticizing link is 0.396.

[0037] The quality control parameter generation unit receives the above two types of coefficients and implements fusion calculation. A two-dimensional decision matrix is established: the X-axis is the quality sensitivity coefficient, and the Y-axis is the parameter deviation coefficient. The plasticizing link inputs (0.644, 0.396), falls into the Ⅲ quadrant (high sensitivity-medium deviation) of the matrix. According to the quadrant rule, the parameter synthesis formula is started: quality control parameter = basic value + sensitivity gain + deviation compensation. The basic value is fixed at 0.5, the sensitivity gain is (0.644-0.5)×0.8=0.115, and the deviation compensation is 0.396×0.3=0.119, and the final quality control parameter is 0.734. The injection link inputs (0.68, 0.28) to obtain the parameter 0.716, and the pressure maintaining link (0.55, 0.31) to obtain the parameter 0.632.

[0038] The multi-dimensional difference analysis unit performs a three-dimensional analysis for each link quality control parameter. The time-domain cumulative deviation analysis focuses on parameter trend: the last 10 minutes of quality control parameter sequence of the plasticizing link is extracted, and the trapezoidal integral method is used to calculate the cumulative deviation. The analysis shows that there is a continuous upward trend in the [09:25, 09:35] interval, and the cumulative amount reaches the warning threshold. The frequency domain energy offset analysis implements fast Fourier transform: after the parameter sequence is converted to the frequency domain, it is detected that the energy proportion of the 0.5-1Hz frequency band is increased to 35% (the historical baseline is 22%), indicating that the abnormal enhancement of the medium frequency fluctuation. The sequence similarity evaluation uses an improved dynamic time warping algorithm: after the current parameter sequence is aligned with the historical normal sequence, the path bending cost is 85 (normal range <60), and the similarity score is only 0.55.

[0039] The difference coefficient matrix generation unit integrates the analysis results. A three-dimensional scoring system is established: time-domain cumulative deviation score = cumulative amount / threshold upper limit (plasticizing link 0.82), frequency domain energy offset score = abnormal frequency band energy proportion (0.35), sequence similarity score = 1-warping cost normalization value (0.45). The final difference coefficient is calculated using the geometric mean: plasticizing link difference coefficient = ∛(0.82×0.35×0.45) = 0.52. Similarly, the injection link difference coefficient is 0.48, and the holding link difference coefficient is 0.41. All results are stored according to the structure of Table 1: Table 1: Injection molding production line difference coefficient matrix.

[0040]

[0041] In the implementation process, the sensitivity coefficient generation uses a distributed cache mechanism to store the process weight factor of the last 100 cycles. The parameter deviation calculation implements real-time calibration: when the historical data sample is insufficient, it automatically switches to cross-line similar process data supplement. The quality control parameter fusion formula sets boundary protection, and the output value is forced to be limited in the [0.2, 0.95] interval to prevent extreme cases. The frequency domain analysis uses sliding window STFT transform, and the window length is automatically adjusted according to the production speed. The matrix generation service is deployed as a high-availability cluster, and the whole production line difference coefficient matrix is refreshed every 30 seconds. The data persistence uses a time series database to store, and retains the complete analysis record of the last 90 days for traceability query. The calculation process implements resource isolation, and allocates an independent calculation container for each production link to avoid mutual interference. When the difference coefficient is detected to be greater than 0.5 for three consecutive times, an early warning event is automatically triggered and pushed to the production monitoring center.

[0042] Example 5: see Figure 5In the implementation process of the injection molding production line, the abnormal source positioning module first constructs a digital network model of the production line by the production topology modeling unit. Based on the physical layout of the production line, each production device is abstracted as a network node, and the node attributes include device type, process parameter interface, upstream and downstream connection relationship, etc. The plasticizer node sets three main parameter interfaces of temperature, pressure and speed, the injection machine node includes injection speed, pressure holding time interface, and the pressure holding machine node configures pressure gradient, cooling rate interface. The connection edge between nodes defines the material flow direction and signal transmission relationship, and the edge from plasticizer to injection machine is marked as melt conveying pipe, and the edge from injection machine to pressure holding machine is defined as mold channel. The model is stored in a graph database, which supports real-time updating of node state and edge weight. When the production line is modified, the network structure is dynamically adjusted through the configuration interface, and the new node automatically inherits the default attribute template of the corresponding link.

[0043] After receiving the difference coefficient matrix from Example 4, the abnormal propagation simulation unit starts the graph-based propagation deduction. The simulation process is divided into three stages: in the initial stage, the difference coefficient is converted into the node infection probability, and the difference coefficient of the plasticizing link is 0.52, which corresponds to the initial infection probability of 65%, the injection link is 0.48, which corresponds to 60%, and the pressure holding link is 0.41, which corresponds to 55%. In the propagation stage, an improved independent cascade model is used, and the propagation probability between adjacent nodes is affected by the type of connection edge: the propagation coefficient of material conveying edge is set to 0.8, and the signal transmission edge is set to 0.5. In each simulation step, the infected nodes try to activate adjacent nodes, and the success or failure of activation is determined by the game result of the propagation coefficient and the resistance of the target node. In the convergence stage, when the number of infected nodes in the whole network changes by less than 5% for three consecutive steps, the simulation is terminated, and the final infection probability of each node is recorded. The simulation results show that the plasticizer node infection probability reaches 82%, the downstream injection machine node is 78%, the pressure holding machine node is 71%, and the non-directly connected temperature control system node accidentally appears 35% infection probability.

[0044] The probability distribution generation unit visualizes the simulation results. A three-dimensional probability heat map is constructed: the X-Y plane maps the physical layout of the production line, and the Z axis represents the infection probability intensity. The plasticizer area presents a deep red high-probability core area, and the probability value decreases radially outward. The heat map is superimposed with a device state layer, which identifies different textures for running, standby, maintenance and other states. The probability distribution data is stored in a space-time database, and each probability point is associated with a timestamp and a data source identifier. The system automatically marks high-risk areas with a probability of more than 75%, and the plasticizer and injection machine connection pipe area is displayed as a red warning belt.

[0045] The physical area positioning unit analyzes the spatial characteristics of the probability distribution map. A density clustering algorithm is used to identify high-probability aggregation areas, and the core radius parameter is set to 1.5 times the average distance between devices. The algorithm outputs three core abnormal areas: the main abnormal area centered on the heating cylinder of the plasticizing machine, the secondary abnormal area of the injection machine nozzle transition area, and the edge abnormal area of the pressure sensor group of the holding machine. Each abnormal area generates polygon boundary coordinates and calculates the mean and dispersion of the probability in the area. The mean probability of the main abnormal area reaches 80%, and the dispersion is less than 15%, indicating that the anomaly is concentrated and stable. The positioning results are converted into device coordinate instructions to control the inspection robot to prioritize the main abnormal area corresponding to the heating cylinder temperature control unit.

[0046] The verification strategy generation unit dynamically configures the detection scheme according to the positioning results. For the core area with a probability of more than 80%, a high-frequency monitoring mode is started: the plasticizing machine temperature sampling frequency is increased from 1 Hz to 10 Hz, and the pressure monitoring is increased from 0.5 Hz to 5 Hz. For the potential risk area with a probability of 35-50%, a node disturbance test is deployed: a ±2°C step signal is injected into the temperature control system node, and the plasticizing machine temperature response delay is monitored. Strategy configuration information is issued to the device controller through the industrial Internet of Things platform, and the execution state is fed back to the central monitoring console in real time. High-frequency monitoring data is stored separately in the cache area, and the last 2 hours of original data are retained for immediate analysis.

[0047] When the system is implemented, the topology modeling uses a what-you-see-is-what-you-get editing interface, supporting drag-and-drop adjustment of device node positions. The propagation simulation introduces a random seed mechanism, generating a difference report for manual review each time. The probability visualization supports multi-view switching, including planar layout view, device profile view, and three-dimensional view. The physical positioning algorithm sets adaptive parameters, automatically reducing the core radius in dense device areas. The verification strategy implementation has priority management, and when multiple abnormal areas conflict, it is processed in descending order of probability value. All operation logs are recorded in the blockchain storage system, including operation time, execution personnel, and result summary.

[0048] The abnormal positioning process implements closed-loop control: after the first positioning result triggers the verification strategy, new evidence found by the verification is fed back to the topology model. When it is found that the temperature control system has a hidden fault, the resistance parameter of its node is updated from 0.6 to 0.4, and after re-simulation, the infection probability of the node rises to 58%. The system continues to iterate until the difference between the two consecutive positioning results is less than 10%, and finally determines that the heating cylinder of the plasticizing machine is the root cause of the anomaly. Throughout the process, resources are dynamically allocated, with computing resources tilted towards analysis tasks in high-probability areas, and lightweight algorithms are used for rapid screening in edge areas.

[0049] Data communication adopts a layered encryption mechanism, with device coordinate instructions encrypted using AES-256 and monitoring data streams transmitted using an SSL channel. System response time is strictly monitored, with the time delay from the difference coefficient matrix input to the generation of the first positioning result controlled within 3 seconds. The user interface provides an interpretable display of the positioning process, with abnormal propagation paths displayed using different color light flow animations and high probability nodes identified using pulse effects. Historical positioning cases are stored in a case library, supporting similarity retrieval functions to assist manual decision-making. When the system version is updated, the new and old versions are compared by running the shadow mode, ensuring smooth transition of the positioning logic changes.

[0050] The maintenance mechanism includes a self-diagnosis function that periodically checks the mapping relationship between the topology model and the actual device. When the injection machine node is found to be out of sync with the physical device, the device re-registration process is automatically triggered. The probability calculation service is deployed as a micro-service cluster, with dedicated computing instances allocated to each production link, and intermediate results exchanged between instances through a low-latency network. The core positioning algorithm implements version control, with both the old and new versions running in parallel for a week before upgrading. Dynamic quotas are implemented for system resource usage, with computing resources automatically released during night shift maintenance for batch historical data analysis. All configuration changes are recorded through a change management system, supporting rollback to any historical version.

[0051] The abnormal positioning results are deeply integrated with the production execution system. After confirming the abnormality of the plasticizing machine heating cylinder, a preventive maintenance work order is automatically triggered, and the scheduling system prioritizes the maintenance period for the device. The intermediate data generated during the positioning process is marked with a time limit, and the simulation results that exceed the valid period are automatically archived to cold storage. The system has multiple levels of permission control, with physical positioning operations requiring permission from process engineers and higher, and verification strategy adjustments requiring authorization from quality managers. All operation instructions are authenticated through digital signatures to ensure traceability of the operation source. The system status panel displays real-time key indicators such as positioning task queue, resource load, and network latency, and triggers a hierarchical alarm mechanism in case of abnormality.

[0052] It should be noted that, in this text, relational terms such as first and second are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "contain" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device that includes a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such a process, method, article or device.

[0053] While embodiments of the application have been shown and described, it is to be understood that the embodiments described are merely exemplary of the principles and application of the present application. Numerous modifications and adaptions can be effected without departing from the spirit and scope of the present application, which is not limited to the exact construction and arrangement described. It is intended, therefore, to cover all modifications and adaptions that fall within the scope of the claims and their equivalents.

Claims

1. A production data layering and quality traceability system, characterized by, The application relates to a production data acquisition module, a data hierarchical processing module, a quality fluctuation index module, a multi-dimensional quality difference analysis module and an abnormal source positioning module. The production data acquisition module is used for collecting real-time parameter data sequences of each production link. The data hierarchical processing module is used for hierarchically processing the real-time parameter data sequences into multiple data levels. The quality fluctuation index module is used for obtaining a quality fluctuation index based on data fluctuation in the multiple data levels. The multi-dimensional quality difference analysis module is used for analyzing multi-dimensional quality differences based on the quality fluctuation index. The abnormal source positioning module is used for positioning a production abnormal source based on the multi-dimensional quality differences. The production data acquisition module comprises a parameter data acquisition unit, a reference data acquisition unit, a temperature sequence acquisition unit, a pressure sequence acquisition unit and a speed sequence acquisition unit.

2. The production data layering and quality traceability system of claim 1, wherein, The reference data acquisition unit is used for collecting historical parameter data sequences of each production link in each production cycle. The data hierarchical processing module comprises an original data layer processing unit, a feature extraction layer processing unit and an aggregation analysis layer processing unit.

3. The production data layering and quality traceability system of claim 2, wherein, The feature extraction layer processing unit is used for extracting a feature parameter sequence from the original data layer processing unit. The aggregation analysis layer processing unit is used for aggregating the feature parameter sequence into a high-order data sequence. The quality fluctuation index module comprises a fluctuation point identification unit, a fluctuation difference calculation unit, a clustering index calculation unit, a mean difference calculation unit and a sudden change index generation unit.

4. The production data layering and quality traceability system of claim 3, wherein, The fluctuation point identification unit is used for identifying each fluctuation point in the feature parameter sequence. The fluctuation difference calculation unit is used for obtaining a forward difference index based on the difference between the fluctuation points. The clustering index calculation unit is used for obtaining a clustering index based on the difference of the feature parameter sequence. The mean difference calculation unit is used for obtaining a pre-post mean difference based on the mean difference of each local feature parameter sequence. The sudden change index generation unit is used for obtaining a quality fluctuation index of each fluctuation point based on the forward difference index, the clustering index and the pre-post mean difference.

5. The production data layering and quality traceability system of claim 4, wherein, The quality fluctuation index module further comprises an average quality fluctuation index generation unit. The average quality fluctuation index generation unit is used for obtaining an average quality fluctuation index of each fluctuation acquisition time of each production cycle based on the average of the quality fluctuation indexes of the fluctuation points.

6. The production data layering and quality traceability system of claim 5, wherein, The average quality fluctuation index generation unit receives the output of the sudden change index generation unit. The quality fluctuation index module further comprises a production abnormal index generation unit.

7. The production data layering and quality traceability system of claim 6, wherein, The production abnormal index generation unit is used for obtaining a production abnormal index of each fluctuation acquisition time of each production cycle based on the distance between local feature parameter sequences and the average quality fluctuation index. The production abnormal index generation unit receives the output of the average quality fluctuation index generation unit. The multi-dimensional quality difference analysis module comprises a quality sensitivity coefficient generation unit and a parameter deviation coefficient generation unit. The quality sensitivity coefficient generation unit is used for obtaining a quality sensitivity coefficient of each production link in each production cycle based on the production abnormal index. The parameter deviation coefficient generation unit is used for obtaining a parameter deviation coefficient based on the deviation of the historical parameter data sequence of the reference data acquisition unit. a quality control parameter generation unit configured to obtain a quality control parameter based on the quality sensitivity coefficient and the parameter deviation coefficient; the input of the quality control parameter generation unit receives the output of the quality sensitivity coefficient generation unit and the parameter deviation coefficient generation unit.

8. The production data layering and quality traceability system of claim 7, wherein, The multi-dimensional quality difference analysis module further includes a multi-dimensional difference analysis unit configured to perform multi-dimensional quality difference analysis based on the quality control parameter, the multi-dimensional quality difference analysis including time-domain cumulative deviation analysis, frequency-domain energy offset analysis, and sequence similarity evaluation; a difference coefficient matrix generation unit configured to generate a difference coefficient matrix based on the time-domain cumulative deviation analysis, the frequency-domain energy offset analysis, and the sequence similarity evaluation; the input of the difference coefficient matrix generation unit receives the output of the multi-dimensional difference analysis unit.

9. The production data layering and quality traceability system of claim 8, wherein, The abnormal source positioning module includes a production topology modeling unit configured to construct a production topology network based on location information of production links; an abnormal propagation simulation unit configured to perform abnormal propagation simulation based on the difference coefficient matrix and the production topology network; a probability distribution generation unit configured to generate an abnormal probability distribution map based on the abnormal propagation simulation; a physical region positioning unit configured to position a production abnormal physical region based on the abnormal probability distribution map; the input of the physical region positioning unit receives the output of the probability distribution generation unit.

10. The production data layering and quality traceability system of claim 9, wherein, The abnormal source positioning module further includes a verification strategy generation unit configured to configure a production verification strategy based on the abnormal probability distribution map, the production verification strategy including high-frequency monitoring mode activation and node disturbance test execution; the input of the verification strategy generation unit receives the output of the physical region positioning unit.

Images