Production process monitoring method and system applied to smart workshop

Abstract

The application provides a production process monitoring method and system applied to a smart workshop, relates to the technical field of industrial production monitoring, and first acquires an initial production process data set containing a process timestamp, a material flow transfer identifier and a station equipment identifier, and performs trend symbolization processing on a process parameter time sequence. Then, an inter-process trend correlation rule library is called to perform correlation consistency analysis, and an abnormal trend symbol unit, an abnormal process and a time are identified. Then, the abnormal root process is determined by reverse tracing, and a production process control instruction is generated and sent to a corresponding station control terminal. The application can accurately monitor the production process in real time, timely adjust and optimize, and improve the production efficiency and quality.

Description

Technical Field

[0001] This invention relates to the field of industrial production monitoring technology, and more specifically, to a production process monitoring method and system applied to smart workshops. Background Technology

[0002] In modern industrial production, smart workshops serve as a key vehicle for improving production efficiency and quality, making efficient monitoring and management of their production processes crucial. Traditional production process monitoring methods mainly rely on regular manual inspections and simple data recording, which have many drawbacks.

[0003] On the one hand, manual inspections struggle to achieve real-time and comprehensive monitoring. Due to the complexity and numerous steps in the workshop production process, it is impossible for personnel to accurately record and analyze the process parameters of all steps at every moment. This can easily lead to the omission of some key information, resulting in the inability to detect potential problems in the production process in a timely manner. For example, in a critical step, process parameters may experience abnormal fluctuations within a very short period of time, but manual inspections may fail to capture this instantaneous change, thus preventing timely handling of the problem and affecting product quality and production efficiency.

[0004] On the other hand, most existing data monitoring methods only analyze the process parameters of individual processes independently, lacking in-depth exploration of the relationships between processes. The various processes in a production flow are not isolated but interconnected and mutually influential. An anomaly in one process may be transmitted to subsequent processes through material flow, triggering a chain reaction. However, traditional methods cannot grasp the overall relationship between these processes, making it difficult to quickly locate the root cause of anomalies, resulting in low problem-solving efficiency and increased production costs and defect rates. Summary of the Invention

[0005] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a production process monitoring method applied to a smart workshop, the method comprising: Acquire the initial production process data set corresponding to the production process in the smart workshop. The initial production process data set includes multiple process parameter time series with process timestamps collected continuously, as well as material flow identifiers and workstation equipment identifiers corresponding to each process parameter time series. The time series of process parameters for each process is processed by trend symbolization. Based on the changing trend of parameter values ​​over time in the time series of process parameters, a trend symbol sequence corresponding to each process is generated. The trend symbol sequence consists of multiple trend symbol units arranged in chronological order. The pre-built inter-process trend association rule library is invoked to perform association consistency analysis on the trend symbol sequences of all processes in the current production process, and to identify abnormal trend symbol units that violate the inter-process trend association rules in the trend symbol sequences, as well as the abnormal process and abnormal time to which the abnormal trend symbol unit belongs. Based on the abnormal process and the abnormal time, reverse tracing is performed in the opposite direction of the material flow direction. The matching degree change characteristics of the trend symbol sequence between adjacent processes are analyzed step by step, and the root cause process of the abnormality is determined based on the matching degree change characteristics. Based on the abnormal root cause process, a production process control instruction containing an abnormal root cause process identifier and a corresponding process parameter adjustment value is generated, and the production process control instruction is sent to the workstation control terminal indicated by the workstation equipment identifier corresponding to the abnormal root cause process to trigger the process parameter adjustment operation.

[0006] Furthermore, embodiments of the present invention also provide a production process monitoring system for smart workshops, comprising: A processor; a machine-readable storage medium for storing machine-executable instructions of the processor; wherein the processor is configured to execute the aforementioned production process monitoring method applied to a smart workshop by executing the machine-executable instructions.

[0007] In another aspect, embodiments of the present invention also provide a computer program product, the computer program product including machine-executable instructions, the machine-executable instructions being stored in a computer-readable storage medium, a processor applied to a production process monitoring system in a smart workshop reading the machine-executable instructions from the computer-readable storage medium, the processor executing the machine-executable instructions, causing the production process monitoring system applied to the smart workshop to execute the aforementioned production process monitoring method applied to the smart workshop.

[0008] Based on the above, by acquiring the initial production process data set within the smart workshop, including process timestamps, material flow identifiers, and workstation equipment identifiers, trend symbolization is performed on the time series of process parameters for each process. This transforms complex parameter changes into intuitive trend symbol sequences, simplifying data analysis while preserving key characteristics of parameter changes. A pre-built inter-process trend association rule library is used to perform consistency analysis on the trend symbol sequences, quickly identifying abnormal trend symbol units that violate the rules and accurately determining abnormal processes and times. Reverse tracing is then performed based on the abnormal processes and times, analyzing the matching degree changes of trend symbol sequences between adjacent processes step by step. This allows for a holistic understanding of the inter-process relationships, accurately locating the root cause process of anomalies, and avoiding misjudgments and omissions caused by isolated analysis of individual processes in traditional methods. Finally, production process control instructions are generated based on the root cause process of anomalies and sent to the corresponding workstation control terminals, enabling timely adjustments and optimizations to the production process. This effectively improves production efficiency, ensures product quality, and reduces production costs and defect rates. Attached Figure Description

[0009] Figure 1 This is a schematic diagram of the execution flow of the production process monitoring method for smart workshops provided in an embodiment of the present invention.

[0010] Figure 2 This is a schematic diagram of exemplary hardware and software components for a production process monitoring system applied to a smart workshop, provided in an embodiment of the present invention. Detailed Implementation

[0011] Figure 1 This is a flowchart illustrating a production process monitoring method for smart workshops provided in one embodiment of the present invention, which will be described in detail below.

[0012] Step S110: Obtain the initial production process data set corresponding to the production process in the smart workshop. The initial production process data set includes multiple process parameter time series with process timestamps collected continuously, as well as material flow identifiers and workstation equipment identifiers corresponding to each process parameter time series.

[0013] In the intelligent textile machine system computing power service project scenario, production process data is acquired through sensors and data acquisition devices deployed at various workstations in the textile workshop. These sensors include yarn tension sensors installed on spinning machines, speed sensors on winding machines, and warp density sensors on weaving machines. They continuously collect the corresponding process parameters at a preset sampling frequency (e.g., once per second), forming multiple process parameter time series. Each data point in the time series is marked with a process timestamp accurate to the millisecond level to accurately record the time nodes of parameter changes. Each process parameter time series is associated with a specific material flow identifier and workstation equipment identifier. The material flow identifier is a unique identifier assigned to each batch of textile raw materials or semi-finished products, and the workstation equipment identifier is used to distinguish different textile equipment, such as spinning machine A1, weaving machine B3, etc. The above identification information, along with the process parameter time series, is stored in a database using a distributed storage architecture on the workshop floor to ensure data reliability and accessibility.

[0014] Step S120: Perform trend symbolization processing on the time series of process parameters for each process. Generate a trend symbol sequence corresponding to each process based on the changing trend of parameter values ​​over time in the time series of process parameters. The trend symbol sequence consists of multiple trend symbol units arranged in chronological order.

[0015] Taking the time series of yarn tension parameters in the spinning process as an example, trend symbolization is performed. First, a comprehensive analysis of the time series is conducted to extract its trend characteristics, because the stability of yarn tension directly affects the quality of subsequent weaving processes, so accurately capturing its changing trend is crucial.

[0016] Step S121: Extract the parameter values ​​of all adjacent time points in the time series of process parameters for each process, calculate the change in parameter values ​​between each adjacent time point, and generate an initial trend symbol sequence based on the positive or negative sign of the change in parameter values. The initial trend symbol sequence contains multiple initial trend symbol units arranged in chronological order, and each initial trend symbol unit corresponds to an adjacent time point pair.

[0017] For the time series of yarn tension parameters in the spinning process, all adjacent time points in the series are traversed. Assuming the yarn tension value collected at time point t1 is F1 and the tension value collected at time point t2 is F2, the change in parameter value between these two adjacent time points is calculated as ΔF = F2 - F1. Initial trend symbol units are generated based on the sign of ΔF: if ΔF > 0, the corresponding initial trend symbol unit is "+", indicating an upward trend in tension; if ΔF < 0, it is "-", indicating a downward trend in tension; if ΔF = 0, it is "0", indicating that the tension remains stable. By performing the above calculation and symbol assignment on all adjacent time point pairs, an initial trend symbol sequence composed of "+", "-", and "0" is obtained. Each symbol unit in this initial trend symbol sequence corresponds one-to-one with its adjacent time point pair, arranged in chronological order.

[0018] Step S122: Perform smoothing filtering on the initial trend symbol sequence, identify trend segments composed of consecutive identical trend symbols in the initial trend symbol sequence, and merge consecutive identical trend symbols into one trend symbol unit to generate a fragmented trend symbol sequence.

[0019] The initial trend symbol sequence may contain high-frequency, small fluctuations caused by sensor noise or brief external interference, which do not represent the true trend of process parameters. Therefore, smoothing filtering is required. A sliding window averaging method is used to process the initial trend symbol sequence, setting a window size containing five consecutive initial trend symbol units. For symbol units within a window, if most symbols are identical, all symbols within that window are merged into the majority symbol. For example, if three "+" and two "-" appear in a window, the window is merged into a "+" window. In this way, trend segments consisting of consecutive identical trend symbols in the initial trend symbol sequence are identified, and each consecutive identical trend symbol is merged into a trend symbol unit, generating a fragmented trend symbol sequence. This fragmented trend symbol sequence more clearly reflects trend changes over a longer time period compared to the initial sequence.

[0020] Step S123: Detect trend turning points in the fragmented trend symbol sequence, extract the positions where trend symbols change in the fragmented trend symbol sequence as trend turning points, and mark the turning point moment for each trend turning point.

[0021] The point where the trend symbol changes within a fragmented trend symbol sequence is the trend inflection point. Inflection points are detected by comparing adjacent trend symbol units within the fragmented trend symbol sequence. A trend inflection point is considered to have occurred when the sequence changes from "+" to "-", or from "-" to "+", or from "+" to "0", from "0" to "+", from "-" to "0", or from "0" to "-". For each detected inflection point, the corresponding inflection time is determined based on its position within the fragmented trend symbol sequence and the original process timestamp. The inflection time is accurate to the time precision corresponding to the original process parameter time series for subsequent analysis and processing.

[0022] Step S124: Divide the fragmented trend symbol sequence into multiple trend segments according to the trend turning point. Each trend segment contains consecutive identical trend symbols. Record the start time, end time, and trend symbol type of each trend segment to generate a trend segment sequence with time period labels.

[0023] After detecting trend inflection points, the fragmented trend symbol sequence is divided into multiple trend segments, using these inflection points as boundaries. Each trend segment consists of consecutive identical trend symbols; for example, a trend segment might consist entirely of "+" symbols, indicating a continuous increase in yarn tension during that time period. For each trend segment, its start and end times are recorded. The start time is the original timestamp corresponding to the first trend symbol unit of the segment, and the end time is the original timestamp corresponding to the last trend symbol unit. The trend symbol type for each segment is also recorded: "+", "-", or "0". This generates a trend segment sequence with time period labels, clearly showing the time range and trend type of each trend change stage.

[0024] Step S125: Perform trend symbol type standardization processing on the trend segment sequence, and map the trend symbol type to the corresponding symbol in the preset standardized trend symbol set. The standardized trend symbol set includes upward trend symbols, downward trend symbols and stable trend symbols.

[0025] To ensure a consistent representation of trend symbols across different processes and parameters, facilitating subsequent correlation analysis, the trend segment sequences are standardized using trend symbol types. The pre-defined set of standardized trend symbols is defined as follows: an upward trend is represented by "U", a downward trend by "D", and a stable trend by "S". "+" in the trend segment sequence is mapped to "U", "-" to "D", and "0" to "S", ensuring that all processes use the same symbol representation and eliminating potential analytical errors caused by inconsistent symbol representations.

[0026] Step S126: For each trend segment, evaluate the time stability represented by its duration and the significance of change represented by the magnitude of its parameter value change; based on the time stability and significance of change, determine the confidence level of the trend segment through preset rules, and store the confidence level with the trend segment to form a trend segment sequence with confidence label.

[0027] For each trend segment, time stability is assessed by the duration of the trend segment; the longer the duration, the more stable the trend and the higher the time stability. Significance of change is measured by the magnitude of change in process parameter values ​​within the trend segment; the greater the magnitude of change, the more significant the trend change. Preset evaluation indicators and grading standards for time stability and significance of change are established. For example, time stability is divided into three levels: high, medium, and low. A duration greater than T1 is considered high stability, between T2 and T1 is medium stability, and less than T2 is low stability. Similarly, significance of change is divided into three levels: high, medium, and low. A change magnitude greater than ΔV1 is high significance, between ΔV2 and ΔV1 is medium significance, and less than ΔV2 is low significance. According to preset rules, the time stability level and significance of change level are combined to determine the confidence level of the trend segment. For example, a trend segment with high stability and high significance has a confidence level of A, medium stability and medium significance is B, and low stability or low significance is C. The confidence level is then associated with the corresponding trend segment and stored to form a sequence of trend segments with confidence labels.

[0028] Step S127: Map the start and end times of each trend segment in the trend segment sequence with confidence markers to the original time axis of the process parameter time series, generate the trend segment identifier and corresponding confidence level of each time point, and form a mapping relationship between time, trend segment and its confidence level.

[0029] Each trend segment in the sequence of trend segments with confidence level labels has a clearly defined start and end time, which are then mapped onto the original timeline. Each point in time on the original timeline corresponds to a process parameter value. By finding the time range of the trend segment to which that point falls, the trend segment identifier and the corresponding confidence level are determined. For example, if a point t on the original timeline falls between the start and end times of trend segment A, then the trend segment identifier for point t is A, and the confidence level is the confidence level of trend segment A, thus forming a mapping relationship between time, trend segment, and its confidence level.

[0030] Step S128: Based on the mapping relationship between the time and the trend segment, regenerate the trend symbol sequence with equal time intervals at a preset time sampling interval, that is, assign the trend symbol type of the trend segment to which the sampling time belongs to each sampling time, and obtain the trend symbol sequence with equal intervals.

[0031] Since the sampling intervals of the original process parameter time series may be inconsistent, or the time lengths of trend segments may vary, a trend symbol sequence with equal time intervals is generated to facilitate subsequent periodic detection and correlation analysis. A fixed time sampling interval is preset, for example, sampling is performed once every T time units. For each sampling moment, based on the mapping relationship between the moment and the trend segment formed in step S127, the trend segment to which the sampling moment belongs is determined. Then, the trend symbol type of the trend segment is assigned to the sampling moment, resulting in an equal-interval trend symbol sequence. Each symbol unit in the equal-interval trend symbol sequence corresponds to a fixed time interval, making the trend symbol sequences of different processes comparable in the time dimension.

[0032] Step S129: Perform periodic detection on the equally spaced trend symbol sequence to identify whether there is a periodically repeating trend symbol pattern in the equally spaced trend symbol sequence. If there is, extract the period length and periodic pattern, and use the periodic pattern as the background trend symbol sequence.

[0033] Many process parameters in textile production exhibit periodicity; for example, the operating cycle of a loom causes periodic changes in warp tension. The autocorrelation function (AFC) method is used to detect the periodicity of equally spaced trend symbol sequences. The AFC measures the similarity of sequences at different time lags; if a sequence exhibits periodicity, the AFC will show a peak at a specific lag value. By calculating the AFC of equally spaced trend symbol sequences, the lag values ​​with larger AFC values ​​are identified. The time lengths corresponding to these lag values ​​are likely the period lengths of the sequences. Based on the period length, corresponding trend symbol patterns are extracted and used as background trend symbol sequences. For example, if the period length of an equally spaced trend symbol sequence is P, then the trend symbol combinations within one period are extracted from the sequence as the periodic pattern, i.e., the background trend symbol sequence.

[0034] Step S1210: Compare the equally spaced trend symbol sequence of each process with the background trend symbol sequence, mark the abnormal trend symbol units that deviate from the background trend symbol sequence, and take the time of the abnormal trend symbol unit as a potential abnormal time candidate.

[0035] After obtaining the equally spaced trend symbol sequence and the corresponding background trend symbol sequence for each process, the two are compared unit by unit. Each symbol unit in the equally spaced trend symbol sequence is compared with the corresponding symbol unit in the background trend symbol sequence in chronological order. If a symbol unit differs from the corresponding symbol unit in the background sequence, it is considered to deviate from the background trend and is marked as an abnormal trend symbol unit. The time of this abnormal trend symbol unit is also recorded as a potential candidate for an abnormal time. For example, if the background trend symbol sequence is "U, U, D, S", and the equally spaced trend symbol sequence is "U, D, D, S" within a certain period, then the second symbol unit "D" differs from "U" in the background sequence, is marked as abnormal, and this time is recorded as a potential candidate for an abnormal time.

[0036] Step S130: Call the pre-built inter-process trend association rule library to perform association consistency analysis on the trend symbol sequences of all processes in the current production process, identify the abnormal trend symbol units that violate the inter-process trend association rules in the trend symbol sequences, as well as the abnormal process and abnormal time to which the abnormal trend symbol unit belongs.

[0037] In intelligent textile machine systems, there are certain correlations between the trends of process parameters in different processes. For example, there may be specific correlation rules between the yarn tension trend in the spinning process and the warp density trend in the weaving process. By calling a pre-built inter-process trend correlation rule library, the correlation consistency analysis of the trend symbol sequences of all processes in the current production process is performed to detect anomalies.

[0038] Step S131: Extract inter-process trend association rules related to all processes in the current production process from the pre-built inter-process trend association rule library. Each inter-process trend association rule includes the trend sign condition of the preceding process, the trend sign condition of the following process, and the time offset constraint between the preceding and following processes.

[0039] The inter-process trend association rule base stores a large number of rules mined from historical production data. Based on the processes involved in the current production flow, relevant rules are extracted from the rule base. Each rule explicitly specifies the trend sign conditions for the preceding process, the trend sign conditions for the succeeding process, and the time offset constraints between them. For example, a rule might stipulate that if the spinning process (the preceding process) shows an upward trend sign "U" at time t, then the weaving process (the succeeding process) should show a stable trend sign "S" at time t+Δt, where Δt is the time offset constraint, representing the time required for the trend change in the spinning process to propagate to the weaving process.

[0040] Step S132: Align the trend symbol sequences of all processes in the current production process according to the time axis to generate a trend symbol matrix with time as the horizontal axis and process as the vertical axis.

[0041] To facilitate correlation analysis of trend symbol sequences across multiple processes, they are aligned on a timeline. Using a common start time as a baseline, the trend symbol sequences of all processes are arranged chronologically, forming a two-dimensional trend symbol matrix. The horizontal axis of the matrix represents time, and the vertical axis represents different processes. Each element in the matrix is ​​the trend symbol of the corresponding process at the corresponding time point. This matrix format visually demonstrates the trend changes of each process over different times.

[0042] Step S133: Traverse the trend symbol matrix in a sliding time window manner, extract the trend symbol unit of each process in each time window, and form the trend symbol vector of the current time window.

[0043] Set a fixed-size sliding time window, for example, containing W time units. Starting from the initial time of the trend symbol matrix, slide the window along the time axis, moving one time unit at a time. At each window position, extract the trend symbol units for all processes within the window. These symbol units are arranged in process order to form a trend symbol vector. This trend symbol vector contains the trend information of each process within the current time window. By analyzing the above vector, it is possible to detect whether the preconditions of the inter-process trend association rule are met.

[0044] Step S134: Match the trend symbol vector with each rule in the inter-process trend association rule base, and check whether the trend symbol vector of the current time window satisfies the previous process trend symbol condition of a certain rule.

[0045] For each extracted trend symbol vector, it is matched against each rule in the rule base. The matching process mainly checks whether the trend symbol corresponding to the preceding process in the vector meets the preceding process trend symbol condition specified in the rule. For example, if the preceding process of a certain rule is spinning, and the trend symbol condition is "U", then the position corresponding to the spinning process is found in the trend symbol vector, and its trend symbol is checked to see if it is "U". If it is, then the preceding condition of the rule is considered to be met; otherwise, it is not.

[0046] Step S135: If the trend symbol condition of the preceding process is satisfied according to any rule, then the expected time window position of the following process is determined according to the time offset constraint of the rule, and the actual trend symbol unit of the following process within the expected time window is extracted.

[0047] When the trend symbol vector satisfies the trend symbol condition of the preceding process according to a certain rule, the expected time window position of the following process is calculated based on the time offset constraint of that rule. For example, if the start time of the current time window is t0 and the time offset constraint is Δt, then the expected start time of the following process's time window is t0 + Δt, and the window size remains W. Then, the actual trend symbol units of the following process within this expected time window are extracted from the trend symbol matrix.

[0048] Step S136: Compare the actual trend symbol unit of the subsequent process within the expected time window with the trend symbol condition of the subsequent process specified in the rule. If the two are inconsistent, it is determined that the rule is violated, and the identifier of the subsequent process that violated the rule, the time of violation, and the identifier of the rule that was violated are recorded.

[0049] The actual trend symbol units of the extracted subsequent processes within the expected time window are compared with the trend symbol conditions for subsequent processes specified in the rules. If the actual trend symbol units are exactly the same as the symbols in the conditions, the rule is satisfied; otherwise, it is determined that the rule is violated. The identifier of the subsequent process that violated the rule, the time of violation (i.e., the start time of the expected time window), and the identifier of the violated rule are recorded for subsequent anomaly analysis.

[0050] Step S137: After traversing all time windows, collect all records that violate the rules and generate an initial set of abnormal records.

[0051] After the sliding time window has traversed the entire trend symbol matrix, all records that violated the rules are aggregated to form an initial set of abnormal records. This initial set of abnormal records contains all cases of violations of the inter-process trend correlation rules, and each record includes information such as the subsequent process identifier, the time of violation, and the rule identifier.

[0052] Step S138: Perform cluster analysis on the initial abnormal record set, group the abnormal records according to the proximity of the violation time and the adjacency of the violation process, and form multiple abnormal event clusters.

[0053] The initial set of anomaly records may be scattered. To better analyze anomalies, cluster analysis is performed. A density-based clustering algorithm, such as DBSCAN, is used to group anomaly records based on the proximity of violation times and the adjacency of violation processes. Records with similar violation times and adjacent violation processes in the production flow are grouped into the same cluster, forming anomaly event clusters. Each anomaly event cluster represents a possible anomaly event.

[0054] Step S139: For each cluster of abnormal events, extract the earliest violation time in the cluster as the start time of the cluster, extract the latest violation time in the cluster as the end time of the cluster, and count the set of processes involved in the cluster.

[0055] For each cluster of abnormal events, the earliest violation time among all abnormal records in the cluster is identified as the start time, and the latest violation time is identified as the end time. At the same time, the processes involved in all abnormal records in the cluster are counted to form a process set. This information helps to determine the time range and impact range of the abnormal event.

[0056] Step S1310: Based on the set of processes involved in the cluster, reverse the derivation of possible abnormal source processes from the inter-process trend association rule base, that is, mark those processes that are the antecedent processes that violate the rules and whose trend symbol conditions are met but cause the subsequent processes to violate the rules as suspected abnormal source processes.

[0057] Based on the set of processes involved within the cluster of anomalies, rules related to these processes are searched from the inter-process trend association rule base. For each rule, if the subsequent process is a process involved within the cluster and the rule is violated, it is checked whether the preceding process of the rule is in the set of processes involved within the cluster and whether its trend sign condition is met. If the trend sign condition of the preceding process is met, but the subsequent process violates the rule, then the preceding process may be the source of the anomaly, and it is marked as a suspected anomaly source process.

[0058] Step S1311: Associate the suspected abnormal source process and its corresponding violation time with the abnormal event cluster to generate a list of suspected abnormal source processes for each abnormal event cluster.

[0059] The marked suspected abnormal source processes and their corresponding violation times are associated with the corresponding abnormal event clusters to form a list of suspected abnormal source processes for each abnormal event cluster. This list of suspected abnormal source processes records the relevant information of each suspected abnormal source process in the abnormal event cluster.

[0060] Step S1312: Use the start time, end time, set of processes involved, and list of suspected abnormal source processes of the abnormal event cluster as the abnormal description information of the abnormal event cluster.

[0061] The start time, end time, set of processes involved, and list of suspected abnormal source processes of an abnormal event cluster are integrated to form the abnormal description information of the abnormal event cluster, which comprehensively describes the basic situation of the abnormal event, including the time range, affected processes, and possible sources.

[0062] Step S1313: Extract the suspected frequency of each suspected abnormal source process from the abnormal description information of all abnormal event clusters, and determine the suspected abnormal source process with the highest suspected frequency as the main suspected abnormal process of the current abnormal event cluster.

[0063] The list of suspected anomaly source processes for all anomaly event clusters is statistically analyzed, and the frequency of each suspected anomaly source process appearing in all clusters is calculated, i.e., the suspected frequency. The suspected anomaly source process with the highest suspected frequency is identified as the primary suspected anomaly process for the current anomaly event cluster. This primary suspected anomaly process is the most likely source of the anomaly event.

[0064] Step S1314: Output the main suspected abnormal process and its earliest violation time as the abnormal process and abnormal time.

[0065] After identifying the main suspected abnormal process, the earliest violation time of the process in the abnormal event cluster is extracted. The main suspected abnormal process is output as the abnormal process and the earliest violation time is output as the abnormal time. This information is used for subsequent abnormal root cause tracing.

[0066] Step S140: Perform reverse tracing processing along the opposite direction of the material flow direction based on the abnormal process and the abnormal time, analyze the matching degree change characteristics of the trend symbol sequence between adjacent processes step by step, and determine the abnormal root cause process based on the matching degree change characteristics.

[0067] In a smart textile machine system, materials flow between processes in a specific order, such as from spinning to winding and then to weaving. After identifying the abnormal process and the time of the abnormality, reverse tracing is performed along the opposite direction of the material flow to find the root cause of the abnormality.

[0068] Step S141: Obtain the abnormal moment of the abnormal process, and determine the traceability path from the abnormal process in the opposite direction of the material flow direction. The traceability path consists of a series of processes that trace back from the abnormal process to the material input process, and there is a material flow relationship between adjacent processes on the traceability path.

[0069] First, identify the moment of the abnormal process. Then, based on the workshop's production layout and material flow relationships, determine a traceability path starting from the abnormal process and moving in the reverse direction of material flow. For example, if the abnormal process is weaving, the preceding process in the reverse direction of material flow might be winding, and the process before that is spinning, and so on, until tracing back to the material input process (such as raw material preparation). Each process on the traceability path has a direct material flow relationship with the preceding process.

[0070] Step S142: Starting from the first process on the tracing path, i.e. the abnormal process, extract the trend symbol subsequence of the abnormal process within a preset time window before and after the abnormal moment, as a reference trend segment of the abnormal process.

[0071] Set a preset time window, such as T1 time units before the abnormal moment and T2 time units after the abnormal moment, to form a time interval containing the abnormal moment. Extract the trend symbol subsequence within this time window from the trend symbol sequence of the abnormal process, as a reference trend segment of the abnormal process. This reference trend segment reflects the trend change of the abnormal process before and after the abnormal moment.

[0072] Step S143: Move forward along the traceability path to the process preceding the abnormal process, and extract the trend symbol subsequence of the preceding process within the time window corresponding to the abnormal moment. The corresponding time window is determined based on the time delay required for the material to flow from the preceding process to the abnormal process.

[0073] Based on the actual production process, there is a certain time delay required for materials to flow from one process to the next. This time delay is obtained through historical production data statistics. For example, if there is a time delay Δt for materials to flow from the winding process to the weaving process, then for the abnormal time t of the abnormal process (weaving process), the time window corresponding to its preceding process (winding process) should be the same size as the time window before and after t-Δt, i.e. (t-Δt-T1, t-Δt+T2). The trend symbol subsequence of the winding process within this time window is extracted.

[0074] Step S144: Calculate the trend matching degree between the trend symbol subsequence of the previous process and the reference trend segment of the abnormal process. The trend matching degree is obtained by comparing the consistency of trend symbols at the same time position in the two subsequences.

[0075] Trend matching degree is used to measure the similarity between two trend symbol subsequences. It involves mapping the trend symbol subsequence of the previous process to the reference trend segment of the abnormal process according to time position, comparing whether the trend symbols at each position are consistent, and counting the proportion of consistent symbols to the total number of symbols. This proportion is used as the trend matching degree. For example, if two subsequences each have N symbols, and M of them are consistent, then the trend matching degree is M / N.

[0076] Step S145: If the trend matching degree exceeds the preset matching threshold, the previous process is marked as a candidate propagation process, and the trend symbol subsequence of the previous process is updated to a new reference trend segment, and the process continues to trace forward.

[0077] The preset matching threshold is determined based on historical data analysis and process requirements, for example, 0.8. If the calculated trend matching degree exceeds this threshold, it indicates that the trend change of the previous process is highly similar to the trend change of the abnormal process, and the abnormality may have been propagated from the previous process. Therefore, the previous process is marked as a candidate propagation process, and its trend symbol subsequence is updated to a new reference trend segment in order to continue tracing back.

[0078] Step S146: If the trend matching degree does not exceed the preset matching threshold, stop tracing forward and mark the previous process of the current process as a candidate abnormal source process.

[0079] If the trend matching degree does not exceed the preset matching threshold, it means that the trend change of the previous process is significantly different from the trend change of the abnormal process, and the abnormality is unlikely to have been propagated from the previous process. At this time, stop tracing forward and mark the previous process of the current process (i.e., the abnormal process or the candidate propagation process) as the candidate abnormal source process. This process may be the root cause of the abnormality.

[0080] Step S147: For all processes marked as candidate propagation processes, record the trend matching degree value between each candidate propagation process and its previous process to form a trend matching degree decay sequence.

[0081] During the reverse tracing process, multiple candidate propagation steps are identified. For each candidate propagation step, the trend matching degree value between it and the previous step is recorded. These matching degree values ​​are arranged in the order of the tracing path to form a trend matching degree decay sequence, which reflects the change of trend matching degree along the tracing path.

[0082] Step S148: Analyze the changing pattern of the trend matching degree decay sequence. If the trend matching degree decay sequence shows a gradual downward trend, then the process preceding the process corresponding to the last matching degree value in the trend matching degree decay sequence is determined as the root cause process of the anomaly.

[0083] If the matching degree value in the trend matching degree decay sequence gradually decreases as the tracing path extends forward, it indicates that the anomaly gradually propagated from a source process to subsequent processes, and the similarity of the trend gradually decreased during the propagation process. In this case, the process corresponding to the last matching degree value in the sequence is the last candidate propagation process, and the process preceding this process is the root cause process of the anomaly.

[0084] Step S149: If the trend matching degree decay sequence shows a pattern of first decreasing and then increasing or fluctuating, then extract the minimum value point in the trend matching degree decay sequence and mark the process preceding the process corresponding to the minimum value point as a potential source of anomalies.

[0085] If the trend matching degree decay sequence is not monotonically decreasing, but instead shows a pattern of first decreasing and then increasing or fluctuating, it indicates that the anomaly propagation process is relatively complex. Extract the minimum point in the sequence; the minimum point corresponds to the lowest matching degree, indicating the greatest difference in process trends before and after this minimum point. Mark the process preceding the process corresponding to the minimum point as a potential source of anomalies.

[0086] Step S1410: For processes marked as potential anomaly source processes, re-extract the trend symbol subsequence of the process within the time window corresponding to the anomaly time, and perform cross-time window matching with the original reference trend segment of the anomaly process to check whether there is a high degree of matching across multiple time windows.

[0087] To further verify whether the potential anomaly source process is the true root cause of the anomaly, trend symbol subsequences for this process are extracted again within different time windows corresponding to the anomaly time. For example, in addition to the previously used time window, subsequences within several time windows with different offsets are extracted. These subsequences are then matched with the original reference trend segment of the anomaly process to check if there are multiple time windows with high matching degrees.

[0088] Step S1411: If there is a high degree of matching for multiple time windows, then the potential source of anomalies is identified as the root cause of anomalies with periodic fluctuation characteristics.

[0089] If the trend symbol subsequence of the potential anomaly source process has a high degree of matching with the original reference trend segment of the anomaly process in multiple different time windows, it indicates that the trend change of the potential anomaly source process is periodic, and the anomaly may be caused by its periodic fluctuations. Therefore, it is identified as the anomaly root source process with periodic fluctuation characteristics.

[0090] Step S1412: If there is no high matching degree across multiple time windows, then take the previous process of the potential anomaly source process as the new tracing starting point and continue tracing forward until the starting point where the trend matching degree shows a monotonically decreasing trend is found, and the process corresponding to the starting point is determined as the root cause process of the anomaly.

[0091] If there is no high matching degree across multiple time windows, it indicates that the potential anomaly source process is not the true root cause of the anomaly. In this case, the process preceding the potential anomaly source process is used as the new starting point for tracing back, and the above reverse tracing process is repeated until the starting point where the trend matching degree decay sequence shows a monotonically decreasing trend is found. The process corresponding to this starting point is the root cause process of the anomaly.

[0092] Step S150: Generate a production process control instruction containing an abnormal root cause process identifier and corresponding process parameter adjustment values ​​according to the abnormal root cause process, and send the production process control instruction to the workstation control terminal indicated by the workstation equipment identifier corresponding to the abnormal root cause process to trigger the process parameter adjustment operation.

[0093] Once the root cause of the anomaly is identified, corresponding production process control instructions are generated to adjust the process parameters of that process in order to eliminate the anomaly.

[0094] Step S151: Analyze the anomaly root cause process identifier and the trend symbol subsequence of the process before and after the anomaly in the anomaly root cause process tracing results.

[0095] Extract the identifier of the abnormal root cause process from the traceability results, such as "spinning process A2", and the trend symbol subsequence of the process before and after the abnormal moment. The minimum value sequence reflects the trend change of the abnormal root cause process before and after the abnormality occurred.

[0096] Step S152: Extract the original process parameter time series segment of the root cause process within the preset time window before the abnormal moment, as the pre-abnormal parameter segment; extract the original process parameter time series segment of the root cause process within the preset time window after the abnormal moment, as the post-abnormal parameter segment.

[0097] Two preset time windows are set. One is the time window before the abnormal moment, for example, time unit T3 before the abnormal moment, from which the original process parameter time series segment is extracted as the parameter segment before the abnormal moment. The other is the time window after the abnormal moment, for example, time unit T4 after the abnormal moment, from which the corresponding segment is extracted as the parameter segment after the abnormal moment. These parameter segments contain the specific process parameter value changes of the root cause process before and after the abnormality occurred.

[0098] Step S153: Input the parameter fragments before and after the anomaly into the pre-trained process parameter adjustment model. The process parameter adjustment model outputs a set of recommended process parameter adjustment values ​​for the root cause process of the anomaly. The set of recommended process parameter adjustment values ​​includes the names and adjustment directions of multiple adjustable process parameters.

[0099] The process parameter adjustment model is an artificial intelligence model trained on historical production data. It is used to recommend appropriate adjustment values ​​based on the changing trends of process parameters. The input of this process parameter adjustment model is parameter fragments before and after the anomaly, and the output is a set of recommended process parameter adjustment values. This set of recommended process parameter adjustment values ​​contains the names of multiple adjustable process parameters, such as "spinning speed" and "yarn tension setpoint", as well as the adjustment direction of each parameter, such as "increase", "decrease" or "remain unchanged".

[0100] Step S154: Select adjustment parameters from the recommended set of process parameter adjustment values ​​that match the current operating mode of the abnormal root cause process. The matching criteria are that the adjustment parameters are compatible with the equipment type of the abnormal root cause process and the adjustment direction conforms to the trend characteristics of the parameter segment after the abnormality.

[0101] Different types of equipment in different processes may have different adjustable process parameters, and the adjustment direction needs to be adapted to the trend characteristics of the parameter segments after the anomaly. For example, if the parameter segments after the anomaly show a continuous upward trend in yarn tension, then the adjustment direction might be to "reduce" the tension set value. Based on the equipment type of the root cause process and the trend characteristics of the parameter segments after the anomaly, the recommended set of process parameter adjustment values ​​is filtered, and matching adjustment parameters are retained.

[0102] Step S155: Encode the selected adjustment parameters and their adjustment directions along with the abnormal root cause process identifier to generate a primary control instruction tuple.

[0103] The selected adjustment parameter names, adjustment directions, and abnormal root cause process identifiers are encoded according to preset coding rules to form primary control instruction tuples. These primary control instruction tuples are the basic data structure of control instructions and contain key information required for adjusting process parameters.

[0104] Step S156: Obtain the current production plan of the production line where the root cause of the anomaly is located, extract the expected output and expected completion time of the root cause of the anomaly from the current production plan, and attach the expected output and expected completion time as control constraints to the primary control instruction tuple.

[0105] Production plans have specific requirements for the output and completion time of each process, and adjustments to process parameters cannot violate these requirements. The expected output and completion time of the root cause process of the anomaly are obtained from the current production plan and added as control constraints to the primary control instruction tuple to ensure that the adjusted process parameters meet the requirements of the production plan.

[0106] Step S157: Verify the feasibility of the adjustment parameters in the primary control instruction tuple according to the control constraints. Verify whether the adjusted process parameters will cause output fluctuations or time delays. If there is a conflict, adjust the magnitude of the adjustment parameters to meet the maximum value of the constraints.

[0107] Based on the control constraints, the feasibility of adjusting the parameters in the primary control instruction tuple is verified. For example, by simulating the impact of the adjusted process parameters on output and production time, it is checked whether this will lead to lower-than-expected output or longer-than-expected completion time. If a conflict exists, the magnitude of the adjustment parameters is adjusted to optimize the process parameters as much as possible while satisfying the constraints, setting the adjustment magnitude to the maximum value that satisfies the constraints.

[0108] Step S158: Convert the primary control instruction tuples that have passed feasibility verification into a data frame format that conforms to the communication protocol of the workstation control terminal, and add the abnormal root cause process identifier and timestamp information to the header of the data frame to generate the final production process control instruction.

[0109] To ensure that control commands can be correctly recognized and executed by the workstation control terminals, the initial control command tuples are converted into data frame formats conforming to a specific communication protocol. Simultaneously, an anomaly root cause process identifier and a current timestamp are added to the data frame header so that the workstation control terminals can confirm the target and execution time of the command. The final generated production process control commands are then sent to the workstation control terminals corresponding to the anomaly root cause process via the workshop's communication network.

[0110] Step S210: The step of pre-constructing the inter-process trend association rule base.

[0111] Step S211: Obtain a historical production process data set, which includes the time series of historical process parameters of multiple normal production batches, as well as the corresponding material flow identifiers and workstation equipment identifiers.

[0112] Collect production data from multiple normal production batches over a past period. This data includes time series of process parameters for each step, material flow identifiers, and workstation equipment identifiers. The historical data should have a sufficiently long time span to cover different production conditions and process variations, ensuring the comprehensiveness and accuracy of the rule base. For example, collect normal production data for different production cycles each month over the past year. Each production batch includes complete process data from raw material input to finished product output. Material flow identifiers correspond to the raw material numbers for each batch, and workstation equipment identifiers correspond to the specific equipment numbers such as spinning machines and weaving machines.

[0113] Step S212: Perform trend symbolization processing on the time series of historical process parameters for each historical production batch to generate a set of historical trend symbol sequences corresponding to each historical production batch.

[0114] The same trend symbolization processing method as in step S120 is used to process the time series of process parameters for each historical production batch. For example, for the yarn tension time series of the spinning process, the parameter changes at adjacent time points are first calculated to generate an initial trend symbol sequence. Then, after steps such as smoothing filtering, inflection point detection, trend segment division, and symbol standardization, a set of historical trend symbol sequences corresponding to each historical production batch is finally obtained. Each historical trend symbol sequence corresponds to one process and one production batch.

[0115] Step S213: Extract all adjacent process pairs from the historical trend symbol sequence set. Each adjacent process pair consists of two processes that have a direct sequential relationship in the material flow direction, and record the process identifier of each adjacent process pair.

[0116] Based on the material flow relationships in the production process, all adjacent process pairs are identified from the historical trend symbol sequence set. For example, in a textile production process, spinning is followed by winding, and winding is followed by weaving; therefore, (spinning, winding) and (winding, weaving) are adjacent process pairs. The process identifier for each adjacent process pair is recorded, such as (spinning A, winding B), (winding B, weaving C), etc., to ensure accurate reflection of the sequential relationships between processes.

[0117] Step S214: For each pair of adjacent operations, align the historical trend symbol sequence of the preceding operation and the historical trend symbol sequence of the following operation according to the time axis, and extract the preceding trend symbol subsequence and the following trend symbol subsequence within all time windows in a sliding time window manner.

[0118] Align the historical trend symbol sequences of the preceding and succeeding processes on the timeline, for example, by using the time when materials begin processing in the succeeding process as a baseline, and shifting the sequence of the preceding process forward by the corresponding material flow time. Then, use a sliding time window to extract sub-sequences. The size of the sliding window is set according to the process characteristics and data sampling frequency, such as containing 5 trend symbol units. For each window position, extract the trend symbol sub-sequences of the preceding and succeeding processes respectively. For example, the sub-sequence of the preceding process is "U, U, D", and the sub-sequence of the succeeding process is "S, U, D".

[0119] Step S215: Calculate the frequency of simultaneous occurrence of the antecedent trend symbol subsequence and the consequent trend symbol subsequence within each time window, and form a co-occurrence frequency matrix of the antecedent trend symbol and the consequent trend symbol.

[0120] All extracted antecedent and consequent trend symbol subsequences are statistically analyzed to calculate the frequency of co-occurrence of different antecedent and consequent trend symbol subsequences. For example, if the antecedent subsequence "U, U, D" and the consequent subsequence "S, U, D" co-occur M times, this frequency is recorded in the corresponding position of the co-occurrence frequency matrix. The rows of the co-occurrence frequency matrix represent the antecedent trend symbol subsequences, the columns represent the consequent trend symbol subsequences, and the matrix elements are the corresponding co-occurrence frequencies. This matrix can visually display the frequency of occurrence of different combinations of trend symbol subsequences.

[0121] Step S216: Calculate the conditional probability of the consequent trend symbol under the condition of the antecedent trend symbol based on the co-occurrence frequency matrix, and determine the trend symbol pairs with conditional probabilities exceeding a preset probability threshold as trend symbol pairs with strong correlation.

[0122] Conditional probability is calculated as follows: Conditional probability = Frequency of simultaneous occurrence of the antecedent and consequent / Total frequency of occurrence of the antecedent. For each antecedent trend symbol subsequence in the co-occurrence frequency matrix, the conditional probability of each consequent trend symbol subsequence is calculated. For example, if the total frequency of the antecedent subsequence "U, U, D" is N, and the frequency of simultaneous occurrence with the consequent subsequence "S, U, D" is M, then the conditional probability is M / N. Trend symbol pairs with conditional probabilities exceeding a preset probability threshold (e.g., 0.7) are identified as trend symbol pairs with a strong correlation. These symbol pairs indicate that after the antecedent trend occurs, the consequent trend is highly likely to follow.

[0123] Step S217: For each pair of trend symbols with a strong correlation, analyze the time difference distribution between the occurrence time of the antecedent trend symbol and the occurrence time of the consequent trend symbol, and extract the main peak position of the time difference distribution as the time offset constraint of the correlation rule.

[0124] For trend symbol pairs with a strong correlation, collect all time differences in their occurrence in historical data, i.e., the time of occurrence of the consequent trend symbol minus the time of occurrence of the antecedent trend symbol. Perform statistical analysis on these time differences and plot a histogram of time difference distribution, with the horizontal axis representing time difference and the vertical axis representing frequency of occurrence. Find the location of the main peak in the distribution; the time difference corresponding to this main peak is the time offset constraint of the association rule. For example, if the time difference distribution shows a significant peak at Δt1, it indicates that after the antecedent trend occurs, the probability of the consequent trend occurring after Δt1 is the highest; therefore, Δt1 is used as the time offset constraint of this rule.

[0125] Step S218: Combine the predecessor process identifier, predecessor trend symbol, successor process identifier, successor trend symbol, and time offset constraint into a candidate process inter-process trend association rule.

[0126] The extracted preceding process identifier, preceding trend symbol subsequence, following process identifier, following trend symbol subsequence, and time offset constraint are combined to form a candidate inter-process trend association rule. For example, a candidate rule can be represented as: the preceding process identifier is spinning process A, the preceding trend symbol is "U, U, D", the following process identifier is winding process B, the following trend symbol is "S, U, D", and the time offset constraint is Δt1. This constitutes a complete candidate rule.

[0127] Step S219: Perform a redundancy check on all candidate process trend association rules. If multiple rules have the same predecessor process, the same predecessor trend symbol, the same successor process, and the same successor trend symbol, then retain the rule with the highest confidence in the time offset constraint.

[0128] Redundancy checks are performed to avoid duplicate or similar rules in the rule base. For multiple rules with the same predecessor process, predecessor trend symbol, successor process, and successor trend symbol, the confidence level of the time offset constraint for each rule is calculated. The confidence level can be measured by indicators such as the peak height or variance of the time difference distribution; the higher the peak and the smaller the variance, the higher the confidence level. The rule with the highest confidence level is retained, and other redundant rules are deleted to ensure the simplicity and effectiveness of the rule base.

[0129] Step S2110: Index the candidate process trend association rules after redundancy check according to the predecessor process identifier and the successor process identifier to build a process trend association rule library.

[0130] The processed candidate rules are categorized and indexed according to the identifiers of the preceding and succeeding processes. For example, a two-dimensional index table can be created, with the row index representing the identifier of the preceding process and the column index representing the identifier of the succeeding process. Each cell stores all the association rules between the preceding and succeeding process pairs. Using this indexing method, relevant rules can be quickly queried and extracted in practical applications, ultimately constructing an inter-process trend association rule library, which is then stored in the system's database.

[0131] Step S310: A step to provide feedback and evaluation on the effects of the production process control instructions after they are executed.

[0132] Step S311: Within the preset evaluation time period after the production process control instruction is sent, continuously collect the process parameter time series of the abnormal root cause process and all its downstream processes to form a set of parameter data after control.

[0133] After the production process control command is sent, a preset evaluation time period is set, such as T5 time units after the control command is executed. The length of this time period is determined based on the stabilization time after process adjustment to ensure that sufficient data reflecting the control effect can be collected. During this time period, the time series of process parameters of the root cause process and all its downstream processes are continuously collected. For example, if the root cause process is spinning process A2, its downstream processes include winding process B3, weaving process C1, etc. The time series of parameters such as yarn tension, speed, and warp density of these processes are collected. These data together constitute the post-control parameter data set, which is used to evaluate the control effect.

[0134] Step S312: Perform trend symbolization processing on the time series of process parameters for each process in the adjusted parameter data set to generate a set of adjusted trend symbol sequences.

[0135] Using the same method as in step S120, trend symbolization is performed on the time series of process parameters for each process in the adjusted parameter data set. Specifically, the parameter changes at adjacent time points are first calculated to generate an initial trend symbol sequence. Then, smoothing filtering, inflection point detection, trend segment division, and symbol standardization are performed to finally obtain the adjusted trend symbol sequence for each process. All these sequences form the adjusted trend symbol sequence set.

[0136] Step S313: Compare the set of trend symbol sequences after regulation with the set of trend symbol sequences in the same time period before regulation, and calculate the similarity of the trend symbol sequences of each process before and after regulation.

[0137] The set of trend symbol sequences after regulation is compared unit by unit with the set of trend symbol sequences for the same time period before regulation (i.e., the time period before and after the anomaly). For example, the trend symbol sequence for spinning process A2 during the anomaly period before regulation is "U, D, S, U", and the sequence for the same time period after regulation is "U, U, S, D". The symbols at each position are compared, and the proportion of identical symbols to the total number of symbols is counted. This proportion is used as the similarity. By calculating the similarity before and after regulation for each process, the impact of regulation on the trend of each process can be assessed.

[0138] Step S314: If the similarity between the trend symbol sequence after the control of the abnormal root cause process and the normal baseline trend symbol sequence exceeds the preset similarity threshold, then the control of the abnormal root cause process is deemed effective.

[0139] The normal baseline trend symbol sequence refers to the trend symbol sequence of the process under normal production conditions, obtained through historical normal production data. For example, the trend symbol sequence of spinning process A2 under normal production conditions is usually "U, U, S, D". The trend symbol sequence after adjustment of the abnormal root cause process is compared with the normal baseline trend symbol sequence, and the similarity is calculated. If the similarity exceeds a preset similarity threshold (e.g., 0.85), it indicates that the trend of the adjusted process has returned to the normal level, and the adjustment is deemed effective.

[0140] Step S315: If the similarity between the adjusted trend symbol sequence of the abnormal root cause process and the normal baseline trend symbol sequence does not exceed the preset similarity threshold, the adjustment is deemed invalid, and the identification step of the abnormal root cause process is re-executed.

[0141] If the similarity does not exceed the preset similarity threshold, it indicates that the control measures have failed to restore the trend of the abnormal root cause process to normal, and the control measures are ineffective. At this time, it is necessary to re-execute steps S130 to S140, that is, to call the inter-process trend association rule base to perform association consistency analysis, perform reverse tracing along the opposite direction of material flow, and other steps to re-identify the abnormal root cause process in order to formulate a new control strategy.

[0142] Step S316: If the control is deemed effective, further analyze the trend symbol sequence of all downstream processes of the abnormal root cause process after the control, and detect whether the trend symbol sequence of the downstream processes has recovered to the normal baseline trend symbol sequence.

[0143] Even if the control of the root cause of the anomaly is effective, downstream processes may not have fully recovered due to the propagation of the previous anomaly. Therefore, it is necessary to analyze the post-control trend symbol sequences of all downstream processes, compare them with their respective normal baseline trend symbol sequences, and check whether the similarity exceeds a preset threshold to determine whether the downstream processes have also recovered to a normal state.

[0144] Step S317: If the trend symbol sequence of all downstream processes returns to normal, it is determined that the abnormal propagation has been blocked and a control success indicator is generated.

[0145] If the similarity between the trend symbol sequences of all downstream processes and the normal baseline trend symbol sequences exceeds a preset threshold, it indicates that the abnormal propagation has been blocked and the entire production process has returned to normal. At this time, a control success flag is generated, and the success information of this control is recorded, including the control time, the root cause process of the abnormality, and the adjusted parameters.

[0146] Step S318: If the trend symbol sequence of some downstream processes has not yet returned to normal, extract the identifier of the downstream processes that have not returned to normal and perform a second reverse tracing to determine whether there is a secondary abnormal source process.

[0147] If the trend symbol sequence of some downstream processes has not returned to normal, it indicates that there may be secondary anomaly source processes. That is, in addition to the identified root cause process, there are other processes causing the downstream processes to be abnormal. Extract the identifiers of these downstream processes that have not returned to normal, such as weaving process C1. Starting from this process, perform a secondary reverse tracing along the opposite direction of material flow. The method is similar to step S140, that is, determine the tracing path, extract the trend symbol subsequence, calculate the trend matching degree, etc., to determine whether there are secondary anomaly source processes.

[0148] Step S319: Generate supplementary control instructions based on the secondary reverse tracing results, and send the supplementary control instructions to the workstation control terminal corresponding to the secondary anomaly source process.

[0149] Based on the secondary anomaly source process identified through secondary reverse tracing, supplementary control instructions are generated according to the method in step S150. This involves parsing the secondary anomaly source process identifier and trend symbol subsequence, extracting parameter fragments before and after the anomaly, inputting them into the process parameter adjustment model to obtain recommended adjustment values, filtering matching adjustment parameters, encoding and generating control instructions, and then sending the supplementary control instructions to the corresponding workstation control terminals to eliminate the secondary anomaly source and restore all downstream processes to normal.

[0150] Step S410: The step of dynamically adapting the inter-process trend association rule base.

[0151] Step S411: Monitor newly added and abandoned process types in the production process in real time. When a new process type is detected, obtain the process parameter definition and material flow relationship of the new process type.

[0152] By interacting in real time with the workshop's production management system, changes in the production process can be monitored. When a new process type is added to the production process, such as the introduction of a new dyeing process, the system can obtain the definition of the process parameters for this new process type, such as the name, unit, and normal range of parameters like dyeing temperature, dyeing time, and dye concentration, as well as its material flow relationship in the production process, i.e., the preceding process (such as the weaving process) and the following process (such as the finishing process), to ensure accurate understanding of the basic information of the new process.

[0153] Step S412: Determine the position of the new process type in the existing process association chain based on the material flow relationship of the new process type, and collect the historical process parameter time series of the new process type during the trial operation.

[0154] Based on the material flow relationship of the new process type, determine its specific position in the existing process chain. For example, the dyeing process is located after the weaving process and before the finishing process. Then, during the trial operation of this process type, collect its process parameter time series. The trial operation period is usually several production batches to ensure that the collected data can reflect the normal operation status of the process.

[0155] Step S413: Perform trend symbolization processing on the historical process parameter time series of the newly added process type to generate the initial trend symbol sequence of the newly added process type.

[0156] The trend symbolization processing method in step S120 is used to process the historical process parameter time series of the newly added process type. For example, for the dyeing temperature time series of the dyeing process, the temperature change at adjacent time points is calculated to generate an initial trend symbol sequence. After smoothing filtering, inflection point detection and other steps, the initial trend symbol sequence of the newly added process type is obtained.

[0157] Step S414: Perform co-occurrence analysis on the historical trend symbol sequences of the newly added process type and its adjacent processes to generate trend association rules between candidate processes related to the newly added process type.

[0158] Following the methods in steps S214 to S218, a co-occurrence analysis is performed on the historical trend symbol sequences of the newly added process type and its adjacent preceding and subsequent processes. Specifically, the time axis is aligned, subsequences within the sliding time window are extracted, co-occurrence frequencies are statistically analyzed, conditional probabilities are calculated, time offset constraints are determined, and trend association rules between candidate processes related to the newly added process type are generated, such as association rules for (weaving process, dyeing process) and (dyeing process, finishing process).

[0159] Step S415: Insert the candidate inter-process trend association rules related to the newly added process type into the existing inter-process trend association rule library and mark them as rules to be verified.

[0160] The generated candidate rules are added to the existing inter-process trend association rule base, and these rules are marked with a "to be verified" tag to distinguish them from the verified mature rules. This allows for the monitoring and verification of these rules during subsequent production processes, ensuring their effectiveness.

[0161] Step S416: Continuously monitor the validity of the rule to be verified during subsequent normal production, count the number of times the rule to be verified is violated, and delete the rule from the rule base if the number of violations exceeds the preset tolerance number.

[0162] After the newly added process type is officially put into production, the execution of the rules to be verified is continuously monitored. During each production process, if the rule to be verified is violated (i.e., the antecedent trend symbol condition is met but the consequent trend symbol condition is not met), a violation count is recorded. When the number of violations exceeds the preset tolerance number (e.g., 5 times), the rule is considered to be inapplicable to the current production conditions or to contain an error, and it is deleted from the rule base to ensure the accuracy of the rule base.

[0163] Step S417: When a discarded process type is detected, delete all rules involving that discarded process type from the inter-process trend association rule library and update the association relationship of the relevant processes.

[0164] If a certain process type is obsolete and no longer used, such as an old spinning process being phased out, delete all rules containing that process type from the rule base, such as predecessor and successor rules related to that spinning process. Simultaneously, update the relationships between other processes related to that process, for example, directly associating the preceding and subsequent processes of that spinning process, ensuring that the relationships between processes in the rule base conform to the current production flow.

[0165] Step S418: Periodically perform global optimization on the entire inter-process trend association rule base, detect whether there are redundant rules, that is, rules with the same predecessor and successor processes but different trend symbol conditions, retain the rule with the highest condition probability and delete the rest.

[0166] Perform global optimization of the rule base periodically (e.g., monthly). For rules with the same predecessor and successor processes but different trend sign conditions, compare their condition probabilities. The higher the condition probability, the stronger the reliability of the rule. Retain the rule with the highest condition probability and delete other redundant rules to improve the efficiency and accuracy of the rule base and reduce the amount of computation during rule matching.

[0167] Step S419: Based on the process adjustments and upgrades in the production process, recalibrate the time offset constraints in the rule base, recalculate the time difference distribution using data from the most recent normal production batch, and update the time offset constraint values.

[0168] When the production process is adjusted or equipment is upgraded, the material flow time between processes may change. For example, replacing conveyor equipment may shorten the material flow time. In this case, using data from the most recent normal production batch, the time difference distribution of the relevant rules is recalculated, that is, the time difference between the time when the trend symbol of the preceding item appears and the time when the trend symbol of the succeeding item appears. The new peak position is found, and the time offset constraint value is updated to ensure the timeliness and accuracy of the rules.

[0169] Step S510: The step of integrating and monitoring multi-source production process data.

[0170] Step S511: Obtain the initial production process data set of multiple parallel production lines in the smart workshop. Each parallel production line corresponds to an initial production process data set, and there are material interaction or resource sharing relationships between the production lines.

[0171] A smart workshop may contain multiple parallel production lines, such as Production Line 1, Production Line 2, etc. These production lines may interact with each other, such as sharing intermediate products (e.g., yarn), or sharing resources, such as sharing dyeing equipment and energy supply. Simultaneously, the initial production process data sets of these parallel production lines are acquired. Each set corresponds to one production line and includes data such as the time series of process parameters, material flow identifiers, and workstation equipment identifiers for all processes on that production line.

[0172] Step S512: Perform the abnormal root cause process identification step on the initial production process data set of each parallel production line to generate the abnormal root cause process identification result for each parallel production line.

[0173] For each parallel production line, steps S110 to S140 are executed independently, namely, acquiring initial data, performing trend symbolization processing, calling the rule base for correlation consistency analysis, and tracing back to determine the root cause of the abnormal process. The root cause of the abnormal process of each production line is identified, and the corresponding identification results are generated. For example, the root cause of the abnormal process of production line 1 is spinning process A2, and the root cause of the abnormal process of production line 2 is spinning process A3, etc.

[0174] Step S513: Summarize the results of identifying the root cause of the abnormality in each parallel production line, and identify the same root cause of the abnormality that occurs simultaneously in different production lines.

[0175] Summarize the results of identifying the root cause of anomalies across all production lines, for example, listing the root cause identifiers for production lines 1 through N. Then check if the same root cause identifier appears simultaneously on different production lines. For example, if the root cause identifier for both production lines 1 and 2 is "spinning process A", it indicates that this process may be experiencing anomalies on multiple production lines.

[0176] Step S514: For the same abnormal root cause process identifier that occurs at the same time, analyze the proximity of the abnormal time of the process in different production lines. If the time difference between the abnormal times is less than the preset time difference threshold, it is determined that the process has cross-line abnormal propagation phenomenon.

[0177] For processes with the same root cause of anomalies occurring simultaneously, compare their anomaly times in different production lines. For example, if the anomaly time for spinning process A in production line 1 is t1, and the anomaly time for spinning process A in production line 2 is t2, calculate the time difference |t1-t2|. If this time difference is less than a preset time difference threshold (e.g., 10 minutes), it indicates that these anomalies may not have occurred independently, but rather that there is cross-line anomaly propagation, meaning the anomaly has spread from one production line to other production lines, possibly due to shared resources or material interactions.

[0178] Step S515: For processes with cross-line anomaly propagation, extract the trend symbol sequence of the process in each production line, analyze the similarity of the trend symbol sequences, and generate a cross-line anomaly pattern description.

[0179] Extract the trend symbol sequences of processes exhibiting cross-line anomaly propagation phenomena in each production line, such as the trend symbol sequences of spinning process A in production lines 1 and 2. Calculate the similarity between these sequences, for example, using the trend matching degree calculation method in step S144, to analyze whether their anomaly patterns are similar, such as whether they both show a trend of first rising and then falling. Based on the analysis results, generate a description of the cross-line anomaly pattern, such as "In production lines 1 and 2, the yarn tension of spinning process A both show a trend of first rapidly rising and then sharply falling, and the anomaly times are close."

[0180] Step S516: Based on the cross-line anomaly pattern description, search for rules involving material interaction or resource sharing between different production lines from the inter-process trend association rule base to determine the path of cross-line anomaly propagation.

[0181] Based on the cross-line anomaly pattern description, rules related to material interactions or resource sharing between different production lines are searched in the inter-process trend association rule base. For example, the rule base may contain rules such as "Spinning process A of production line 1 supplies yarn to winding process B of production line 2," or "Production line 1 and production line 2 share dyeing equipment C." These rules can be used to determine the path of anomaly propagation between production lines, such as how the shared dyeing equipment causes anomalies to spread from production line 1 to production line 2.

[0182] Step S517: Mark all processes on the cross-line anomaly propagation path as a set of processes that require coordinated control, and generate corresponding process parameter adjustment values ​​for each process in the set.

[0183] Based on the determined cross-line anomaly propagation path, all processes along the path are marked as a set of processes requiring coordinated control. For example, if the propagation path is spinning process A in production line 1 → shared dyeing equipment C → weaving process D in production line 2, then the set of processes requiring coordinated control includes spinning process A, dyeing equipment C, and weaving process D. For each process in the set, corresponding process parameter adjustment values ​​are generated according to the method in step S150, such as adjusting the yarn tension of spinning process A, the temperature of dyeing equipment C, and the warp density of weaving process D.

[0184] Step S518: Arrange the process parameter adjustment values ​​of each process according to the order of cross-line anomaly propagation path to generate a multi-production line collaborative control instruction set.

[0185] To ensure effective control, the adjustment values ​​of process parameters for each process are arranged according to the sequence of the cross-line anomaly propagation path. For example, the spinning process A at the source of the propagation is adjusted first, then the dyeing equipment C in the middle is adjusted, and finally the downstream weaving process D is adjusted, generating a multi-production line coordinated control instruction set. This ensures that the adjustments of each process are carried out in the order of anomaly propagation, avoiding the inability to effectively eliminate anomalies due to improper adjustment order.

[0186] Step S519: Send the multi-production line collaborative control instruction set to the workstation control terminal of the corresponding process of each production line, and monitor the execution status of each terminal to ensure the synchronization of collaborative control.

[0187] The coordinated control command set is sent to the corresponding workstation control terminals, such as the spinning process A control terminal and dyeing equipment C control terminal in production line 1, and the weaving process D control terminal in production line 2. The execution status of each terminal is monitored in real time, including whether the command has been received, started, and completed. Through timestamp synchronization and other methods, it is ensured that the processes of each production line can be adjusted synchronously according to the programmed sequence, effectively preventing the propagation of anomalies across lines and restoring all production lines to normal operation.

[0188] For example, step S610: the step of visualizing the monitoring results of the production process.

[0189] Step S611: Collect the time series of process parameters, trend symbol sequence, abnormal root cause process identification results, and control command execution status of each process in real time to form a monitoring data stream.

[0190] The system collects various data in real time during the production process, including time series of process parameters for each step (such as yarn tension in spinning and warp density in weaving), processed trend symbol sequences (such as "U, D, S, U"), results of identifying the root causes of anomalies (such as anomaly process identifiers and anomaly times), and the execution status of control commands (such as sent, in progress, completed, and failed). This data is then integrated into a monitoring data stream in chronological order to ensure its real-time performance and completeness.

[0191] Step S612: Organize the monitoring data stream according to the production line and process to generate a process flow monitoring view with the time axis as the horizontal axis and the process as the vertical axis. Each process in the process flow monitoring view corresponds to a time sequence trajectory.

[0192] The monitoring data stream is organized according to the hierarchical structure of production lines and processes. For example, data is first divided by production line, and then arranged sequentially within each production line according to the process order. A process flow monitoring view is generated in the visualization interface. The horizontal axis of the view is the time axis, representing the passage of time from left to right; the vertical axis is the process name, arranged from top to bottom according to the production flow sequence. Each process corresponds to a time-series trajectory in the view. Changes in the shape and color of the trajectory reflect the trend of process parameter changes in that process; for example, an upward trajectory indicates an increase in parameter value, and a downward trajectory indicates a decrease in parameter value.

[0193] Step S613: In the process flow monitoring view, mark the trend symbol type of each process at the current moment with different colors or shapes to form a trend symbol dynamic coloring layer.

[0194] To visually display the trend symbol types for each process, a dynamic coloring layer for the trend symbol is added to the timeline of each process in the process flow monitoring view. For example, red represents an upward trend symbol "U", blue represents a downward trend symbol "D", and green represents a stable trend symbol "S". As time progresses, the coloring layer dynamically changes according to the real-time trend symbol type, highlighting the current trend symbol type with the corresponding color or shape, allowing operators to quickly understand the current trend status of each process.

[0195] Step S614: Overlay the abnormal root cause process mark in the process flow monitoring view, mark the identified abnormal root cause process with a highlighted icon, and attach a floating prompt with the abnormal time and abnormal root cause process identifier.

[0196] Once the root cause of the anomaly is identified, a highlighted icon, such as a flashing red triangle or exclamation mark, is overlaid on the corresponding process sequence in the process flow monitoring view. When the operator hovers the mouse over the icon, a hover tooltip displaying the time of the anomaly and the identifier of the root cause process will appear, such as "Abnormal Process: Spinning Process A2, Abnormal Time: 2023-10-20 14:30:00", allowing operators to quickly understand the specific information of the anomaly.

[0197] Step S615: Draw arrows indicating the direction of material flow between processes in the process flow monitoring view, and connect strongly related process pairs with special lines according to the rules in the process trend association rule library to form a rule association visualization layer.

[0198] To illustrate the material flow relationships and trend correlation rules between processes, material flow direction arrows are drawn in the process flow monitoring view. These arrows point from upstream to downstream processes, indicating the direction of material flow, such as from the spinning process to the winding process. Simultaneously, process pairs with strong correlations in the inter-process trend correlation rule library are connected by special lines (such as bold orange lines). These lines can be labeled with information such as the time offset constraints of the rules, forming a rule correlation visualization layer to help operators understand the inherent connections and trend influences between processes.

[0199] Step S616: Set up a control instruction panel below or to the side of the process flow monitoring view to display the content, target process, and execution status of the most recently sent production process control instructions in real time.

[0200] Below or to the side of the process flow monitoring view, a control command panel is set up. This control command panel uses a list format to display the detailed content of the most recently sent production process control commands in real time, including the name of the adjusted process parameter (such as "yarn tension setpoint"), the adjustment direction (such as "reduce"), the target process identifier (such as "spinning process A2"), and the execution status of the command (such as "executed", "in execution", "execution failed"). Operators can intuitively understand the implementation status and effect of control measures through this panel.

[0201] Step S617: When the user clicks the highlighted icon of a certain process, a detailed monitoring information window for that process pops up. The detailed monitoring information window includes the historical curve of the trend symbol sequence of the process, parameter segments before and after the abnormal moment, and the history of related control instructions.

[0202] When an operator clicks the highlighted icon of a specific process in the process flow monitoring view, a detailed monitoring information window pops up. The window is divided into several areas. The left area displays the historical trend symbol sequence curve of that process, visually showing the trend changes over a period of time in a line graph format. The middle area displays process parameter segments before and after the anomaly, such as a table showing parameter value changes 10 minutes before and after the anomaly. The right area lists the history of control commands related to that process, including command content, sending time, and execution results, helping operators to deeply analyze the causes of the anomaly and the effectiveness of control measures.

[0203] Step S618: When a user selects a trend association rule between processes, the preceding and succeeding processes involved in the rule are highlighted in the process flow monitoring view, and a schematic animation of the time offset constraint is dynamically displayed.

[0204] Operators can select and view trend correlation rules between processes from the system's rule list. When a rule is selected, the system highlights the preceding and succeeding processes involved in the rule in the process flow monitoring view, for example, by highlighting their timing trajectories with a yellow border, and drawing an arrow connecting the two processes. Time offset constraints are displayed through dynamic animation, such as drawing a time-marked arrow between the location where the trend symbol appears in the preceding process and the expected location where the trend symbol appears in the succeeding process, and using the movement or color change of the arrow to indicate the time delay relationship, helping operators understand the specific meaning of the rule and the time relationship.

[0205] Step S619: Periodically save snapshots of the process flow monitoring view and key indicators of the monitoring data stream as monitoring reports. The monitoring reports include statistics on the frequency of anomalies in each process, statistics on the success rate of control, and record of rule base updates.

[0206] The system periodically (e.g., daily or weekly) saves snapshots of the process flow monitoring view. These snapshots visually reflect production flow trends and anomalies within that timeframe. Simultaneously, it performs statistical analysis on key indicators of the monitoring data stream, generating monitoring reports. These reports include statistics on the frequency of anomalies for each process (e.g., monthly anomaly count, anomaly rate), control success rate statistics (e.g., the percentage of successful controls out of total controls, average control time), and rule base update records (e.g., number of added rules, number of deleted rules, rule version numbers). These reports help managers understand the overall operational status of the production process and the effectiveness of the system's monitoring.

[0207] In one exemplary embodiment, a production process monitoring system for a smart workshop is provided. This system can be a terminal, a server, etc., and its internal structure diagram can be as follows: Figure 2As shown, the production process monitoring system for a smart workshop includes a processor, memory, input / output interfaces, a communication interface, a display unit, and input devices. The processor, memory, and input / output interfaces are connected via a system bus, and the communication interface, display unit, and input devices are also connected to the system bus via the input / output interfaces. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides the environment for the operation of the operating system and computer programs in the non-volatile storage media. The input / output interfaces are used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, near-field communication, or other technologies. When the computer program is executed by the processor, it implements a production process monitoring method for a smart workshop. The display unit is used to form a visually visible image and can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device can be a touch layer covering the display screen, or a button, trackball, or touchpad set on the shell of a production process monitoring system used in a smart workshop, or an external keyboard, touchpad, or mouse, etc.

[0208] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. A production process monitoring method applied to a smart workshop, characterized in that, The method includes: Acquire the initial production process data set corresponding to the production process in the smart workshop. The initial production process data set includes multiple process parameter time series with process timestamps collected continuously, as well as material flow identifiers and workstation equipment identifiers corresponding to each process parameter time series. The time series of process parameters for each process is processed by trend symbolization. Based on the changing trend of parameter values ​​over time in the time series of process parameters, a trend symbol sequence corresponding to each process is generated. The trend symbol sequence consists of multiple trend symbol units arranged in chronological order. The pre-built inter-process trend association rule library is invoked to perform association consistency analysis on the trend symbol sequences of all processes in the current production process, and to identify abnormal trend symbol units that violate the inter-process trend association rules in the trend symbol sequences, as well as the abnormal process and abnormal time to which the abnormal trend symbol unit belongs. Based on the abnormal process and the abnormal time, reverse tracing is performed in the opposite direction of the material flow direction. The matching degree change characteristics of the trend symbol sequence between adjacent processes are analyzed step by step, and the root cause process of the abnormality is determined based on the matching degree change characteristics. Based on the abnormal root cause process, a production process control instruction containing an abnormal root cause process identifier and a corresponding process parameter adjustment value is generated, and the production process control instruction is sent to the workstation control terminal indicated by the workstation equipment identifier corresponding to the abnormal root cause process to trigger the process parameter adjustment operation.

2. The production process monitoring method applied to a smart workshop according to claim 1, characterized in that, The step of performing trend symbolization processing on the time series of process parameters for each process step, and generating a trend symbol sequence corresponding to each process step based on the changing trend of parameter values ​​over time in the time series of process parameters, includes: Extract the parameter values ​​of all adjacent time points in the time series of process parameters for each process, calculate the change in parameter values ​​between each adjacent time point, and generate an initial trend symbol sequence based on the positive or negative sign of the change in parameter values. The initial trend symbol sequence contains multiple initial trend symbol units arranged in chronological order, and each initial trend symbol unit corresponds to an adjacent time point pair. The initial trend symbol sequence is smoothed and filtered to identify trend segments composed of consecutive identical trend symbols in the initial trend symbol sequence, and consecutive identical trend symbols are merged into a trend symbol unit to generate a fragmented trend symbol sequence. The fragmented trend symbol sequence is subjected to trend turning point detection. The positions where the trend symbols in the fragmented trend symbol sequence change are extracted as trend turning points, and the turning point is marked for each trend turning point. Based on the trend turning point, the fragmented trend symbol sequence is divided into multiple trend segments. Each trend segment contains consecutive identical trend symbols, and the start time, end time, and trend symbol type of each trend segment are recorded to generate a trend segment sequence with time period labels. The trend segment sequence is subjected to trend symbol type standardization processing, and the trend symbol type is mapped to the corresponding symbol in the preset standardized trend symbol set, which includes upward trend symbols, downward trend symbols and stable trend symbols. For each trend segment, the time stability represented by its duration and the significance of change represented by the magnitude of its parameter value change are evaluated respectively. Based on the time stability and significance of change, the confidence level of the trend segment is determined by a preset rule, and the confidence level is associated with the trend segment and stored to form a sequence of trend segments with confidence labels. The start and end times of each trend segment in the trend segment sequence with confidence labels are mapped to the original time axis of the process parameter time series, generating the trend segment identifier and corresponding confidence level of each time point, thus forming a mapping relationship between time, trend segment and its confidence level. Based on the mapping relationship between the time and the trend segment, a trend symbol sequence with equal time intervals is regenerated at a preset time sampling interval. That is, each sampling time is assigned the trend symbol type of the trend segment to which that time belongs, and an equal interval trend symbol sequence is obtained. Periodic detection is performed on the equally spaced trend symbol sequence to identify whether there is a periodically repeating trend symbol pattern in the equally spaced trend symbol sequence. If there is, the period length and period pattern are extracted, and the period pattern is used as the background trend symbol sequence. The equally spaced trend symbol sequence of each process is compared with the background trend symbol sequence to mark the abnormal trend symbol units that deviate from the background trend symbol sequence, and the time of the abnormal trend symbol unit is taken as a potential abnormal time candidate.

3. The production process monitoring method applied to a smart workshop according to claim 1, characterized in that, The process involves calling a pre-built inter-process trend association rule base to perform association consistency analysis on the trend symbol sequences of all processes in the current production flow, identifying abnormal trend symbol units that violate the inter-process trend association rules, as well as the abnormal process and abnormal time to which the abnormal trend symbol unit belongs, including: Extract inter-process trend association rules related to all processes in the current production process from the pre-built inter-process trend association rule library. Each inter-process trend association rule includes the trend sign condition of the preceding process, the trend sign condition of the following process, and the time offset constraint between the preceding and following processes. Align the trend symbol sequences of all processes in the current production process according to the time axis to generate a trend symbol matrix with time as the horizontal axis and process as the vertical axis. The trend symbol matrix is ​​traversed in a sliding time window manner, and the trend symbol units of each process in each time window are extracted to form the trend symbol vector of the current time window. The trend symbol vector is matched with each rule in the inter-process trend association rule base, and it is checked whether the trend symbol vector of the current time window satisfies the previous process trend symbol condition of a certain rule. If the trend sign condition of the preceding process is met according to any rule, the expected time window position of the following process is determined according to the time offset constraint of the rule, and the actual trend sign unit of the following process within the expected time window is extracted. Compare the actual trend symbol unit of the subsequent process within the expected time window with the trend symbol condition of the subsequent process specified in the rule. If the two are inconsistent, it is determined that the rule is violated, and the identifier of the subsequent process that violated the rule, the time of violation, and the identifier of the rule that was violated are recorded. After iterating through all time windows, collect all records that violate the rules and generate an initial set of exception records; Cluster analysis is performed on the initial set of abnormal records, and the abnormal records are grouped according to the proximity of the violation time and the adjacency of the violation process to form multiple abnormal event clusters; For each cluster of abnormal events, the earliest violation time in the cluster is extracted as the start time of the cluster, and the latest violation time in the cluster is extracted as the end time of the cluster. The set of processes involved in the cluster is also counted. Based on the set of processes involved in the cluster, the possible abnormal source processes are deduced from the inter-process trend association rule base. That is, those processes that are the antecedent processes that violate the rules and whose trend sign conditions are met but cause the subsequent processes to violate the rules are marked as suspected abnormal source processes. Associate the suspected abnormal source process and its corresponding violation time with the abnormal event cluster to generate a list of suspected abnormal source processes for each abnormal event cluster; The start time, end time, set of processes involved, and list of suspected abnormal source processes of an abnormal event cluster are used as the abnormal description information of the abnormal event cluster. Extract the suspected frequency of each suspected anomaly source process from the anomaly description information of all anomaly event clusters, and determine the suspected anomaly source process with the highest suspected frequency as the main suspected anomaly process of the current anomaly event cluster. The main suspected abnormal process and its earliest violation time are output as the abnormal process and abnormal time.

4. The production process monitoring method applied to a smart workshop according to claim 1, characterized in that, The step of performing reverse tracing processing along the opposite direction of material flow based on the abnormal process and the abnormal time, analyzing the matching degree change characteristics of the trend symbol sequence between adjacent processes step by step, and determining the root cause process of the abnormality based on the matching degree change characteristics, includes: The abnormal moment of the abnormal process is obtained, and a traceability path is determined from the abnormal process in the opposite direction of the material flow. The traceability path consists of a series of processes that trace back from the abnormal process to the material input process, and there is a material flow relationship between adjacent processes on the traceability path. Starting from the first process on the tracing path, i.e. the abnormal process, extract the trend symbol subsequence of the abnormal process within a preset time window before and after the abnormal moment, as a reference trend segment of the abnormal process. Move forward along the traceability path to the process preceding the abnormal process, and extract the trend symbol subsequence of the preceding process within the time window corresponding to the abnormal moment. The corresponding time window is determined based on the time delay required for the material to flow from the preceding process to the abnormal process. Calculate the trend matching degree between the trend symbol subsequence of the previous process and the reference trend segment of the abnormal process. The trend matching degree is obtained by comparing the consistency of trend symbols at the same time position in the two subsequences. If the trend matching degree exceeds the preset matching threshold, the previous process is marked as a candidate propagation process, and the trend symbol subsequence of the previous process is updated to a new reference trend segment, and the process continues to trace forward. If the trend matching degree does not exceed the preset matching threshold, then stop tracing forward and mark the previous process of the current process as a candidate abnormal source process; For all processes marked as candidate propagation processes, record the trend matching degree value between each candidate propagation process and its predecessor process to form a trend matching degree decay sequence; Analyze the changing pattern of the trend matching degree decay sequence. If the trend matching degree decay sequence shows a gradual downward trend, then the process preceding the process corresponding to the last matching degree value in the trend matching degree decay sequence is identified as the root cause process of the anomaly. If the trend matching degree decay sequence shows a pattern of first decreasing and then increasing or fluctuating, then the minimum point in the trend matching degree decay sequence is extracted, and the process preceding the process corresponding to the minimum point is marked as a potential source of anomalies. For processes marked as potential anomaly sources, the trend symbol subsequence of the process within the time window corresponding to the anomaly time is re-extracted and matched across time windows with the original reference trend segment of the anomaly process to check whether there is a high degree of matching across multiple time windows. If there is a high degree of matching among multiple time windows, then the potential source of anomalies is identified as the root cause of anomalies with periodic fluctuation characteristics. If there is no high matching degree across multiple time windows, the process preceding the potential source of anomaly is taken as the new starting point for tracing, and the tracing continues forward until the starting point where the trend matching degree shows a monotonically decreasing trend is found. The process corresponding to the starting point is then identified as the root cause of the anomaly.

5. The production process monitoring method applied to a smart workshop according to claim 2, characterized in that, The method also includes the step of pre-constructing a rule base for inter-process trend association, specifically including: Acquire a historical production process data set, which includes the time series of historical process parameters of multiple normal production batches, as well as the corresponding material flow identifiers and workstation equipment identifiers. For each historical production batch, the time series of historical process parameters are processed to generate a set of historical trend symbol sequences corresponding to each historical production batch; Extract all adjacent process pairs from the set of historical trend symbols. Each adjacent process pair consists of two processes that have a direct sequential relationship in the material flow direction, and record the process identifier of each adjacent process pair. For each pair of adjacent operations, the historical trend symbol sequence of the preceding operation and the historical trend symbol sequence of the following operation are aligned according to the time axis, and the preceding trend symbol subsequence and the following trend symbol subsequence within all time windows are extracted in a sliding time window manner. The frequency of simultaneous occurrence of the antecedent trend symbol subsequence and the consequent trend symbol subsequence within each time window is counted to form a co-occurrence frequency matrix of the antecedent trend symbol and the consequent trend symbol. The conditional probability of the consequent trend symbol under the condition of the antecedent trend symbol is calculated based on the co-occurrence frequency matrix. Trend symbol pairs with conditional probabilities exceeding a preset probability threshold are identified as trend symbol pairs with strong correlation. For each pair of trend symbols with a strong correlation, analyze the time difference distribution between the occurrence time of the antecedent trend symbol and the occurrence time of the consequent trend symbol, and extract the main peak position of the time difference distribution as the time offset constraint of the correlation rule. Combine the preceding process identifier, preceding process trend symbol, following process identifier, following process trend symbol, and time offset constraint into a single trend association rule between candidate processes. Redundancy checks are performed on all candidate process trend association rules. If multiple rules have the same predecessor process, the same predecessor trend symbol, the same successor process, and the same successor trend symbol, the rule with the highest confidence in time offset constraint is retained. After redundancy checks, the inter-process trend association rules are indexed according to the predecessor process identifier and the successor process identifier to construct an inter-process trend association rule library.

6. The production process monitoring method applied to a smart workshop according to claim 3, characterized in that, The step of generating a production process control instruction containing an abnormal root cause process identifier and corresponding process parameter adjustment values ​​based on the abnormal root cause process includes: Analyze the anomaly root cause process identifier and the trend symbol subsequence of the process before and after the anomaly moment in the anomaly root cause process tracing results; Extract the original process parameters time series segment of the root cause process within a preset time window before the abnormal moment, as the parameter segment before the abnormality; Extract the original process parameters time series fragments within a preset time window after the abnormality time of the root cause process, and use them as post-abnormal parameter fragments; The parameter fragments before and after the anomaly are input into a pre-trained process parameter adjustment model. The process parameter adjustment model outputs a set of recommended process parameter adjustment values ​​for the root cause of the anomaly. The set of recommended process parameter adjustment values ​​includes the names and adjustment directions of multiple adjustable process parameters. The recommended process parameter adjustment value set is used to select adjustment parameters that match the current operating mode of the abnormal root cause process. The matching criteria are that the adjustment parameters are compatible with the equipment type of the abnormal root cause process and the adjustment direction conforms to the trend characteristics of the parameter segment after the abnormality. The selected adjustment parameters and their adjustment directions are encoded along with the abnormal root cause process identifier to generate a primary control instruction tuple. Obtain the current production plan of the production line where the root cause of the anomaly is located, extract the expected output and expected completion time of the root cause of the anomaly from the current production plan, and attach the expected output and expected completion time as control constraints to the primary control instruction tuple. The feasibility of adjusting parameters in the primary control instruction tuple is verified based on the control constraints. It is verified whether the adjusted process parameters will cause output fluctuations or time delays. If there is a conflict, the magnitude of the adjustment parameters is adjusted to the maximum value that satisfies the constraints. The initial control instruction tuples, after feasibility verification, are converted into data frame format that conforms to the communication protocol of the workstation control terminal. An abnormal root cause process identifier and timestamp information are added to the header of the data frame to generate the final production process control instructions.

7. The production process monitoring method applied to a smart workshop according to claim 4, characterized in that, The method also includes a step of feedback evaluation of the effects of the execution of production process control instructions, specifically including: Within a preset evaluation period after the production process control instruction is sent, the time series of process parameters of the root cause of the abnormality and all its downstream processes are continuously collected to form a set of parameter data after control. The time series of process parameters for each process in the adjusted parameter data set are processed to generate a set of adjusted trend symbol sequences. The set of trend symbol sequences after regulation is compared with the set of trend symbol sequences in the same time period before regulation, and the similarity of the trend symbol sequences before and after regulation for each process is calculated. If the similarity between the trend symbol sequence after the control of the abnormal root cause process and the normal baseline trend symbol sequence exceeds a preset similarity threshold, then the control of the abnormal root cause process is deemed effective. If the similarity between the trend symbol sequence after the control of the abnormal root cause process and the normal baseline trend symbol sequence does not exceed the preset similarity threshold, the control is deemed invalid, and the identification step of the abnormal root cause process is re-executed. If the control is deemed effective, further analyze the trend sign sequence of all downstream processes of the abnormal root cause process after the control, and detect whether the trend sign sequence of the downstream processes has recovered to the normal baseline trend sign sequence. If the trend symbol sequence of all downstream processes returns to normal, it is determined that the abnormal propagation has been blocked, and a control success indicator is generated. If the trend symbol sequence of some downstream processes has not yet returned to normal, the identifier of the downstream processes that have not returned to normal is extracted and a second reverse tracing is performed to determine whether there are secondary abnormal source processes. Based on the secondary reverse tracing results, supplementary control instructions are generated and sent to the workstation control terminal corresponding to the secondary anomaly source process.

8. The production process monitoring method applied to a smart workshop according to claim 5, characterized in that, The method further includes a step of dynamically adapting the inter-process trend association rule base, specifically including: Real-time monitoring of newly added and abandoned process types in the production process; when a new process type is detected, the process parameter definition and material flow relationship of the new process type are obtained. Based on the material flow relationship of the new process type, determine its position in the existing process association chain, and collect the historical process parameter time series of the new process type during the trial operation; Perform trend symbolization on the historical process parameter time series of the newly added process type to generate the initial trend symbol sequence of the new process type; Co-occurrence analysis is performed on the historical trend symbol sequences of the newly added process type and its adjacent processes to generate trend association rules among candidate processes related to the newly added process type. Insert the candidate inter-process trend association rules related to the newly added process type into the existing inter-process trend association rule library and mark them as rules to be verified; During subsequent normal production, the validity of the rules to be verified is continuously monitored, and the number of times the rules to be verified are violated is counted. If the number of violations exceeds the preset tolerance number, the rule is deleted from the rule base. When a discarded process type is detected, all rules involving that discarded process type are deleted from the inter-process trend association rule base, and the association relationships of related processes are updated. Regularly perform global optimization of the entire inter-process trend association rule base, detect whether there are redundant rules, that is, rules with the same predecessor and successor processes but different trend symbol conditions, retain the rule with the highest condition probability and delete the rest; Based on the process adjustments and upgrades in the production process, the time offset constraints in the rule base are recalibrated, and the time difference distribution is recalculated using data from the most recent normal production batches to update the time offset constraint values.

9. The production process monitoring method applied to a smart workshop according to claim 6, characterized in that, The method also includes a step of fusing and monitoring multi-source production process data, specifically including: Acquire the initial production process data set of multiple parallel production lines in the smart workshop. Each parallel production line corresponds to an initial production process data set, and there are material interaction or resource sharing relationships between the production lines. For each parallel production line, an abnormal root cause process identification step is performed on the initial production process data set to generate abnormal root cause process identification results for each parallel production line. The results of identifying the root cause of anomalies in each parallel production line are summarized, and the same root cause of anomalies that occur simultaneously in different production lines are identified. For the same abnormal root cause process identifier that appears at the same time, analyze the proximity of the abnormal time of the process in different production lines. If the time difference between the abnormal times is less than the preset time difference threshold, it is determined that the process has cross-line abnormal propagation phenomenon. For processes exhibiting cross-line anomaly propagation, extract the trend symbol sequence of the process in each production line, analyze the similarity of the trend symbol sequences, and generate a cross-line anomaly pattern description. Based on the description of cross-line anomaly patterns, rules involving material interaction or resource sharing between different production lines are searched from the inter-process trend association rule base to determine the path of cross-line anomaly propagation. Mark all processes on the cross-line anomaly propagation path as a set of processes that require coordinated control, and generate corresponding process parameter adjustment values ​​for each process in the set; The process parameter adjustment values ​​of each process are arranged in the order of cross-line anomaly propagation path to generate a multi-production line collaborative control instruction set; The multi-production line collaborative control instruction set is sent to the workstation control terminal of the corresponding process on each production line, and the execution status of each terminal is monitored to ensure the synchronization of collaborative control.

10. A production process monitoring system applied to a smart workshop, characterized in that, include: processor; A machine-readable storage medium for storing machine-executable instructions of the processor; The processor is configured to execute the production process monitoring method for a smart workshop as described in any one of claims 1 to 9 by executing the machine-executable instructions.

Images