Electromechanical industry digital process evaluation method based on data analysis

Abstract

The invention discloses an electromechanical industry digital process evaluation method based on data analysis, and particularly relates to the technical field of digital evaluation, a behavior path map, a time sequence causal dependence model and an operation frequency density model are constructed based on information provided by the electromechanical industry, whether structural deviation or behavior abnormity exists is analyzed, and the evaluation result is obtained. If the initial deviation label is generated, calculating a behavior consistency score and a process integrity score, inputting the scores into a fuzzy logic device for risk reasoning, and outputting a credible level; according to the credibility level, judging whether the data is suitable for digital process evaluation, and if the data does not meet the preset requirement, feeding back a corresponding optimization strategy; according to the method, digital behavior structured analysis is realized by constructing a behavior path map, a time sequence causal dependence model and an operation frequency density model; a deviation triggering mechanism is set, and abnormal operation behaviors are accurately recognized; the credibility level is output based on the scoring result, the information applicability is judged according to the credibility level, an optimization strategy is fed back, and the evaluation intelligence level is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of digital evaluation, and more specifically, to a method for evaluating the digitalization process of the electromechanical industry based on data analysis. Background Art

[0002] With the continued advancement of the Industrial Internet, intelligent manufacturing, and digital transformation strategies, the electromechanical industry is accelerating its evolution toward a highly information-based and automated future. In practical applications, enterprises deploying manufacturing execution systems (MES), supervisory control and data acquisition (SCADA), and intelligent terminal interaction platforms enable real-time access to extensive user operation records, equipment status logs, and process execution data from production lines. This data provides a foundation for evaluating and optimizing digital processes, offering broad coverage, fine temporal granularity, and diverse information types.

[0003] Existing digital evaluation schemes for the electromechanical industry usually rely on static indicators, such as equipment operating rate, product qualification rate, task completion rate, etc. Although these result-oriented indicators can reflect macro trends, they cannot capture micro-behavioral characteristics such as abnormal states in the behavior process, process deviations in the execution path, or short-term behavior fluctuations, making it difficult to make reliable judgments on the authenticity and integrity of the data.

[0004] In the process of analyzing process execution and user behavior in existing technologies, there is a common tendency to misjudge the operation behavior as "structural compliance is trustworthy", ignoring the abnormal phenomenon of behavior fluctuation in the time dimension. For example, although some operation paths are consistent with the process in terms of node structure, Figure 1 However, its behavioral characteristics such as intensive operations, abnormal increase in the number of events or uneven distribution of response time in a short period of time are very likely caused by data supplementation, script injection or non-real operations, thus affecting the authenticity of the data and the accuracy of process evaluation. This type of "operation density anomaly" cannot be directly identified through process structure diagrams or sequence rules. It is a hidden problem with good structural matching but abnormal behavioral characteristics. The existing methods lack effective modeling means and reliable judgment basis, which seriously restricts the construction of a high-reliability digital evaluation system. Therefore, the present invention proposes a method for evaluating the digital process of the electromechanical industry based on data analysis, in order to solve the above problems. Summary of the Invention

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A method for evaluating the digitalization process of the electromechanical industry based on data analysis includes the following steps:

[0007] User operation records, equipment status logs, and process execution data are extracted from information provided by the electromechanical industry. Through structural analysis, a behavior path map, a temporal causal dependency model, and an operation frequency density model are established. The behavior path map describes the operation path between users and equipment, the temporal causal dependency model reflects the process execution sequence and logical constraints, and the operation frequency density model describes the behavior fluctuation characteristics within a unit of time.

[0008] Perform analysis and matching to determine whether to generate an initial deviation label. If an initial deviation label has been generated, perform further evaluation: Calculate a behavioral consistency score based on the degree of path deviation in the behavioral path map and the degree of fluctuation anomaly in the operation frequency density model. Calculate a process integrity score based on the logical chain breakpoints and dependency conflict degree in the temporal causal dependency model.

[0009] The behavior consistency score and process integrity score are used as inputs to a machine learning model pre-trained with historical data from the electromechanical industry to perform risk reasoning and output the corresponding trust level of the information provided by the current electromechanical industry.

[0010] The following judgments are made based on the credibility level:

[0011] The information provided by the electromechanical industry meets the preset trust requirements and is suitable for the final digital process evaluation, and the digital process evaluation is carried out according to the normal process;

[0012] The information provided by the electromechanical industry does not meet the preset trust requirements and is not suitable for the final digital process evaluation. Based on the trust level that does not meet the preset trust requirements, a corresponding information optimization strategy is proposed and fed back to the electromechanical industry.

[0013] In a preferred embodiment, the construction of the behavior path map includes the following process:

[0014] Sort the user operation records and device status logs in the information provided by the electromechanical industry in chronological order, extract the operation time, operation subject, operation action and device response information, and generate user behavior graph nodes and device response graph nodes;

[0015] Directed graph edges are constructed between graph nodes in the form of action-response pairs to form a continuous interaction path between the user and the device. The behavior path graph is a set of behavior sequence graph structures with sequential constraints. Each path is initiated by a specific user and connects multiple device response graph nodes in series according to time logic.

[0016] In a preferred embodiment, the construction of the temporal causal dependency model includes the following process:

[0017] Pre-process the process execution data in the information provided by the electromechanical industry, treat each process as a process dependency node, extract the start time, end time, and task trigger conditions of each process, and construct a process dependency table based on the execution sequence in the process flow;

[0018] A directed acyclic graph structure is generated based on the dependency table. The direction of the edge in the graph represents the process dependency sequence. Each process dependency node contains a time stamp and a trigger condition field to determine whether the process sequence is violated.

[0019] In a preferred embodiment, the construction of the operation frequency density model includes the following process:

[0020] User operation records and device status logs are divided into time periods of equal length. The total number of corresponding operation events in each time period is calculated to generate a time period operation point sequence. Each time period operation point records the number of operation events and operation category distribution information within the time period. An operation density curve is constructed based on the changing trend of the number of operation events, and local extreme values, slope mutation points, and change rate information are extracted from the curve to jointly reflect the intensity of behavioral fluctuations.

[0021] In a preferred embodiment, performing analysis and matching and determining whether to generate an initial deviation label refers to:

[0022] The operation paths in the behavior path map are matched with the process nodes in the temporal causal dependency model to determine whether the interactive behavior has order reversal, step skipping, or response lag relative to the process logic. At the same time, based on the sharp fluctuations in the number of operation events in the corresponding time period in the operation frequency density model, it is determined whether there is a density anomaly. If the match fails or the fluctuation anomaly reaches the preset anomaly threshold, an initial deviation label is generated based on the corresponding anomaly type.

[0023] In a preferred embodiment, the logic for obtaining the behavior consistency score is:

[0024] The path deviation value is calculated based on the behavior path graph. The path deviation value is obtained by counting the number of mismatched behaviors, key node jumps, and hysteresis operations in the interactive behavior path relative to the process logic, and dividing this total by the total number of operation steps defined in the standard path to obtain a normalized ratio reflecting the degree of structural deviation.

[0025] Fluctuation anomalies are calculated based on the operation frequency density model. This is done by dividing user operation records and device status logs into multiple time periods of equal length. The magnitude of the change in the number of operation events within each time period is then counted. The absolute value of the difference in the number of operation events between adjacent time periods is processed and averaged. If a time period experiences a sudden change in the number of events, a local extreme value, or an abnormal slope of change, a weight greater than one is assigned to that time period to increase its impact on the overall fluctuation.

[0026] The path deviation value and the fluctuation anomaly value are input into the rule-penalized scoring structure together. The path deviation value is squared to enhance the penalty intensity of structural anomalies, and the fluctuation anomaly value is squared to smooth out the high-frequency mutation anomalies, thereby generating a behavioral consistency score, which is a continuous real number between zero and one.

[0027] In a preferred embodiment, the logic for obtaining the process integrity score is:

[0028] The process chain breakage rate is calculated based on the temporal causal dependency model. The process chain breakage rate is calculated by counting the number of process dependencies that fail to be triggered in the interactive behavior path and dividing this number by the total number of dependencies in the predefined process flow. This reflects the proportion of process chains that are not activated in the interactive behavior.

[0029] Calculate the process conflict rate. The process conflict rate is calculated by identifying the number of process relationships that violate the dependency direction in the interactive behavior path and dividing this number by the total number of dependency relationships in the process flow. It reflects the proportion of behaviors in which the execution order is reversed or the dependency conditions are violated in actual execution.

[0030] Calculate the process coverage correction term. The process coverage correction term is formed by counting the number of process nodes successfully executed in the actual behavior path and dividing this number by the total number of nodes defined in the process to form a compensation factor to measure the completeness of the process execution;

[0031] The process chain breakage rate and process conflict rate are converted linearly and nonlinearly respectively, and the process coverage correction term is introduced to form the scoring function. The specific method is: multiply the process chain breakage rate by the first weight parameter, multiply the square root of the process conflict rate by the second weight parameter, and use the sum of the above two items to construct the basic scoring value in the form of one minus one, and then multiply it by the process coverage correction term to form the final process integrity score. The scoring value is a continuous real number between zero and one.

[0032] In a preferred embodiment, the machine learning model is implemented by a fuzzy logic controller, and the reasoning method of the fuzzy logic controller includes the following steps:

[0033] Obtain the behavior consistency score and process integrity score, and input them into the fuzzy logic device as input variables; then perform fuzzification on the two input variables according to multiple preset segmentation thresholds, and divide the input variables into three fuzzy sets of low level, medium level and high level; then set the output variable as the credibility level of the information provided by the current electromechanical industry, and divide the output variable into three levels of high credibility, medium credibility and low credibility, forming a fuzzy set of output variables; formulate fuzzy rules based on the relationship between the input variable level combination and the output level, perform fuzzy reasoning operations on the fuzzified input variables according to the preset fuzzy rules, and obtain the fuzzy membership result corresponding to the output credibility level, and defuzzify the output variable based on the maximum membership principle, and finally output the credibility level corresponding to the information provided by the current electromechanical industry.

[0034] The technical effects and advantages of the present invention are as follows:

[0035] This paper extracts user operation records, equipment status logs, and process execution data from information provided by the electromechanical industry, and constructs a behavioral path map, a temporal causal dependency model, and an operation frequency density model, respectively. This approach constructs a structured modeling approach for three dimensions: user behavior, process flow, and operation density. This structured processing method can organize the raw operation information in a temporal order and restore its logic, thus unifying behavioral data from different sources. This significantly improves the ability to analyze digital behavioral processes, providing a data foundation and structural support for subsequent behavioral matching and trustworthy assessment.

[0036] This paper incorporates an analytical matching mechanism that uses behavioral path graphs and temporal causal dependency models to match operation sequences with process nodes. It also analyzes behavioral fluctuations in conjunction with an operation frequency density model, thereby identifying issues such as sequence reversals, step skipping, response lags, and density anomalies. By determining whether the conditions for generating initial deviation labels are met, a behavioral screening mechanism based on structural and frequency anomalies is implemented. This avoids redundant calculations across all data and performs scoring analysis only on behavioral paths with potential anomalies, improving system processing efficiency and recognition accuracy.

[0037] This method uses behavioral consistency scores and process integrity scores as inputs into a trained machine learning model for risk inference. It then outputs a confidence level, further determining the suitability of the current information for the final digital process evaluation and proposing corresponding information optimization strategies. This mechanism implements a closed-loop logic process, from structural analysis to risk assessment to outcome determination. This effectively enhances the automation, intelligence, and feedback capabilities of digital process evaluation, helping electromechanical companies promptly identify problematic data and optimize operational processes, ultimately improving overall data quality and process stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings;

[0039] Figure 1 This is a schematic diagram of a method for evaluating the digitalization process of the electromechanical industry based on data analysis in the present invention. DETAILED DESCRIPTION

[0040] The following will provide a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0041] Reference Figure 1 The following examples were obtained:

[0042] Example 1: A method for evaluating the digitalization process of the electromechanical industry based on data analysis, comprising the following steps:

[0043] By extracting user operation records, equipment status logs, and process execution data from information provided by the electromechanical industry and performing structured analysis on these raw operation data, three behavioral modeling systems with different functional roles were constructed. Specifically, user operation records, equipment status logs, and process execution data were extracted from information systems and parsed into data structures suitable for path analysis, process dependency analysis, and density analysis, respectively, using unified encoding of timestamps, operation subjects, and response objects. First, a behavioral path graph was constructed, establishing directed operation-response relationships between nodes in the user behavior graph and nodes in the device response graph to describe the continuous interaction between users and devices. Second, a temporal causal dependency model was constructed, defining each process as a process node and constructing a directed acyclic dependency graph structure based on the precedence constraints in the process logic to reflect the sequential execution relationships of the digital process. Third, an operation frequency density model was constructed, dividing user operations and equipment logs into unified time periods. The operation frequency and category distribution within these time periods were statistically analyzed to form a density curve with a time series structure, which was used to measure the changing characteristics of the operation rhythm. The collaborative construction of these three models provided data support and a structured analysis foundation for subsequent behavioral anomaly identification and credibility calculation.

[0044] Perform analysis and matching to determine whether to generate an initial deviation label. Based on the three types of models constructed, perform node and path mapping analysis between the behavioral path graph and the temporal causal dependency model, and combine the time period density fluctuation characteristics in the operation frequency density model to determine whether there are signs of behavioral anomalies. Path mapping analysis is used to detect whether interactive behaviors conform to process logic, including but not limited to whether the order of operations is reversed, whether there are key step skips, and whether there are response lags. The density model is used to identify whether operational behaviors experience abnormal fluctuations within certain time periods, such as a sharp increase or decrease in event density or local extreme values. If a behavioral path contains a path pattern that does not conform to the logical structure, or if the density fluctuation exceeds the set threshold, the current behavior is deemed abnormal, the data segment is marked as a deviation sample, and an initial deviation label is generated. This label serves as the entry condition for the subsequent scoring process, ensuring that only behavioral paths with potential anomalies enter the deep credibility assessment, saving computing resources and focusing on key risk behaviors.

[0045] When behavioral data, or information provided by the electromechanical industry, is flagged as containing initial deviations (i.e., generating an initial deviation label), the system automatically initiates the calculation of behavioral consistency and process integrity scores. The behavioral consistency score is jointly modeled using two metrics: path structure deviation and behavioral fluctuation intensity. It reflects the overall consistency and rhythmic stability of user-device interaction behavior. The process integrity score is calculated using three metrics: process chain breakage rate, execution sequence conflict rate, and coverage node ratio. It reflects the degree to which the behavioral path adheres to the process structure. The behavioral consistency and process integrity scores are fed into a machine learning model pre-trained using historical electromechanical industry data for risk inference, outputting a trustworthiness rating corresponding to the current information provided by the electromechanical industry. Based on the scoring results, the two scores are then fed as input features into a machine learning model pre-trained using historical industrial data for risk inference. This model, constructed using a fuzzy logic processor, fuzzifies the two scores and performs fuzzy rule inference on them. It then outputs a trustworthiness rating label that represents the overall trustworthiness of the current behavioral data across the dimensions of structure, sequence, and behavior intensity. Through this inference process, raw, complex data is transformed into clear, quantifiable, and predictable trustworthiness ratings, providing a technical basis for digital decision-making.

[0046] The following judgments are made based on the trust level: if it meets the preset trust requirements, the information provided by the electromechanical industry is suitable for the final digital process evaluation, and the digital process evaluation is carried out according to the normal process; if it does not meet the preset trust requirements, the information provided by the electromechanical industry is not suitable for the final digital process evaluation, and the corresponding information optimization strategy is proposed based on the trust level that does not meet the preset trust requirements and is fed back to the electromechanical industry. After completing the trust level judgment, the system classifies the information currently provided by the electromechanical industry. If the scoring result meets the preset trust level standard, it is judged that the data segment has sufficient authenticity and integrity, and can be used as an effective input for the subsequent digital process evaluation and enter the regular evaluation process; if the scoring result does not meet the trust standard, it means that there is a significant structural or behavioral anomaly in the current behavior path, and the data segment will not be adopted for evaluation. At the same time, the system derives corresponding information optimization suggestions based on the detailed features that do not meet the level, and feeds back to the data provider.

[0047] The construction of a behavioral path map involves the following steps: First, the user operation records and device status logs from the information provided by the electromechanical industry are sorted in chronological order. User operation records refer to operational behavior data entered or executed by enterprise personnel through human-computer interaction interfaces, industrial terminals, industrial control systems, and other means, and include fields such as the initiator of the operation, the time of the operation, and the content of the operation. Device status logs refer to the real-time responses, event changes, and state transitions of the device during operation, which are usually recorded automatically by the device and include fields such as the device number, response time, and the type of response event. By jointly sorting these two types of data by time field, we ensure that the operational behaviors and device responses are arranged in an orderly manner along the timeline, providing a foundation for the subsequent establishment of the behavioral graph structure.

[0048] Extract the operation time, operation subject, operation action and device response information. The operation time is used to mark the temporal position of the event. The operation subject refers to the person, position or terminal identification who performs the operation. The operation action describes the specific behavioral intention (such as "start", "pause", "parameter setting", etc.). The device response information includes the device identification code and the corresponding response behavior (such as "motor start feedback", "sensor status change"). Based on this information, construct the user behavior graph node and the device response graph node. The user behavior graph node represents the operation initiated by the user at a certain point in time, and the device response graph node represents the response behavior generated by the device within the corresponding time period.

[0049] Directed graph edges are constructed between graph nodes in the form of action-response pairs, indicating a direct logical and temporal dependency between the action and the device response. Directed edges originate at user action graph nodes and end at device response graph nodes, indicating that the device response was triggered or influenced by the user action. If multiple device responses are logically associated with the same user action, directed edges can be established from the user node to the multiple device response nodes. This ultimately forms a continuous interaction path between the user and the device. The behavior path graph is a set of behavior sequence graphs with sequential constraints. Each path is initiated by a specific user and connects multiple device response graph nodes in a temporal logical manner. Sequential constraints require that actions must conform to a natural order of occurrence and must not skip key steps or be executed in reverse order. Each path represents a typical action process instance, consisting of an action starting point, multiple intermediate action-response interaction nodes, and a possible termination state. The behavior path graph fully reconstructs the real-time interaction relationship between the user and the device in a structured manner, providing a logical path basis for subsequent process matching, deviation identification, and scoring analysis.

[0050] For example, in a workshop's automated feeding control system, a user sends the "Start feeding" command through the human-machine interface at 08:01. This command is recorded as a user operation record. Subsequently, the feeding motor returns to the "Started" state at 08:01:05, and the controller records "Signal response successful" at 08:01:08, resulting in two device status logs. The system arranges these three data points in chronological order, extracts the time, user ID, operation action, and device event, and generates a user behavior graph node and two device response graph nodes, constructing the operation path on both sides of the three nodes. Ultimately, this path is added to the behavior path map for subsequent path structure comparison and consistency scoring with the process flow.

[0051] The construction of a temporal causal dependency model involves the following steps: First, preprocessing the process execution data provided by the electromechanical industry. Process execution data refers to the actual execution status of each process step in the manufacturing process, including data fields such as process number, operation time, execution status, and task relationships. Preprocessing includes operations such as standardizing time formats, removing abnormal records, and filling in logically missing fields to construct a structured process model.

[0052] Each process is considered a process dependency node. A process dependency node is a structural unit in the model that represents a single process step. It has a unique process identifier and can contain multiple associated attributes. For each process dependency node, its start time, end time, and task trigger condition are extracted. The start time and end time respectively mark the start and completion time of the process during actual execution, which is used to determine the temporal sequence. The task trigger condition indicates whether the process depends on the completion of other preceding processes or the receipt of external instructions before it can be started. For example, "Process B can only be started when process A is completed and the sensor status is active."

[0053] Based on the above information, a process dependency table is constructed based on the execution sequence within the process flow. A process flow is a set of standard execution steps defined according to product design and manufacturing process standards. Each step has strict sequential logic and conditional constraints. The process dependency table is a two-dimensional relationship matrix or mapping table that structurally describes this sequential logic, listing the dependency types and triggering conditions between each process and its upstream processes.

[0054] A directed acyclic graph (DAG) is generated based on the dependency table. Each process dependency node in the graph is represented by a node element, and nodes are connected by directed edges. Directed edges point from upstream process nodes to downstream process nodes, indicating that the upstream process must complete before the downstream process can proceed. This acyclic nature ensures that there are no logical loops in the entire process flow, preventing execution deadlocks or task loops caused by dependency conflicts.

[0055] Each process dependency node contains a time stamp and trigger condition fields, which are used to determine whether the process sequence is violated. During the process analysis, if it is found that the actual execution time of a process is earlier than its logical upstream process, or the trigger condition is not met before execution, it can be identified as a process logic conflict or sequence violation. The core goal of this model is to construct a set of logical flow charts with complete dependency description capabilities, which will be used for subsequent structural and timing comparison with the actual operation paths in the behavior path map to identify whether there are broken links, sequence errors or dependency conflicts in the execution process, thereby providing basic support for process integrity scoring.

[0056] Taking an automated assembly line as an example, products must sequentially undergo four processes: "Material Loading → Position Calibration → Automatic Screw Locking → Pass Inspection." Preprocessing unifies the execution timeframes of each process into a standard time format, and extracts the task trigger condition for the "Material Loading" process as "Material Sensor Status: In Place." After constructing the dependency table, "Material Loading" becomes the first process node, with a directed edge pointing to the "Position Calibration" node, which in turn connects to "Automatic Screw Locking" and "Pass Inspection." If actual logs reveal that "Automatic Screw Locking" is executed prematurely before "Position Calibration" is complete, this is considered a process execution conflict, and the relevant node will be identified as a logical anomaly in the process integrity score.

[0057] The construction of the operation frequency density model includes the following process: First, divide the user operation records and equipment status logs into time periods of equal length. User operation records refer to behavioral data initiated by personnel, terminals or programs in a digital operating system, which contain information such as the time of operation initiation, operation type, and execution target; equipment status logs refer to state change data automatically generated by production equipment during operation, usually including fields such as timestamp, status code, event type, and result tag. To achieve unified analysis, these two types of data will be segmented and processed based on a unified time starting point and time period length, and divided into multiple time periods of equal length. Time period division can be set at minutes, seconds, or custom granularity to form a time window sequence for statistics.

[0058] Within each time period, the system compiles statistics on all operation and device event records that occurred within that time period, calculates the total number of operation events corresponding to that time period, and generates a sequence of time period operation points. Time period operation points are the core structural unit of the model, representing the frequency of operations within each time window. Each time period operation point records the number of operation events and the distribution of different operation types within that time period. Operation categories can be categorized and coded based on operation objectives, functional categories, and permission levels, allowing the model to simultaneously reflect both the number of actions and the structural characteristics of these actions.

[0059] An operation density curve is constructed based on the changing trend of the number of operation events. This curve is a time series curve formed by sequentially connecting a set of operation points over a certain time period. The vertical axis represents the total number of operation events, and the horizontal axis represents the order of the time periods. This curve reflects the temporal distribution of operation behavior during digital execution and can be considered a mapping of operation intensity over time.

[0060] After the operation density curve is established, the local extreme values, slope mutation points and change rate information in the curve are further extracted. Local extreme values ​​are used to mark the high-frequency peaks or troughs of operation behaviors; slope mutation points indicate the locations where the operation frequency increases or decreases sharply between adjacent time periods; and the change rate measures the average fluctuation in the number of operation events per unit time. These indicators will serve as a structural characterization of the intensity of fluctuations in operation behaviors by the model, and will be used to identify whether there are short-term high-intensity operations, concentrated reporting of behaviors, abnormal operation rhythms, and other phenomena. The final generated operation frequency density model will be represented in the form of structured data, providing data sources and technical support for the calculation of fluctuation anomalies in subsequent behavior consistency scores.

[0061] For example, an automated assembly line in a CNC workshop operated for ten minutes between 9:00 and 9:10, divided into ten one-minute time periods. During the three time periods of 9:01, 9:02, and 9:03, the number of recorded operation events in each period was 5, 24, and 6, respectively. The operation category at 9:02 was concentrated in the "debugging command" category. This situation manifests itself in the density curve as a peak in the middle segment, with the slope increasing from 5 to 24 before plummeting to 6. The system identifies this trend as a sudden slope change and a local extreme value, marking it as an abnormal density segment. This density anomaly is subsequently weighted as a fluctuation anomaly in the behavioral consistency score, significantly impacting the score.

[0062] Performing analysis and matching, and deciding whether to generate an initial deviation label, involves matching the action paths in the behavior path graph with the process nodes in the temporal causal dependency model. A behavior path graph is a graph structure that describes the continuous interaction between a user and a device. Its action paths are sequential paths consisting of user behavior graph nodes and device response graph nodes. The process nodes in the temporal causal dependency model represent the various standard process units in the process flow. Directed dependencies exist between nodes, indicating the order of execution.

[0063] In the corresponding matching process, first match the behavior node timestamp and device response content in each operation path in the map to the process node position it should belong to in the flowchart. Through sequence comparison, dependency verification and trigger condition identification, determine whether the operation behavior is correctly executed according to the process logic. If there is an inconsistency with the process node in the behavior path, it specifically includes the following three situations: one is the order reversal, that is, the subsequent process is triggered in advance when the previous process has not been completed; the second is step jumping, that is, the intermediate process is not executed and jumps to the subsequent process; the third is response lag, that is, there is an abnormal delay in the response time of the corresponding equipment after the operation behavior is completed. Any of the above situations is regarded as an inconsistency between the interactive behavior and the process logic.

[0064] The operation frequency density model determines whether there are density anomalies based on sharp fluctuations in the number of operation events within a corresponding time period. The operation frequency density model divides user operation records and device status logs into time periods of equal length and calculates the number of operation events and the distribution of operation categories within each time period to construct an operation density curve. If the number of operation events within a certain time period shows an extreme increase or a sudden decrease relative to adjacent time periods, and this change exceeds the preset anomaly threshold, it is considered a density anomaly. Such anomalies may reflect issues such as concentrated user operations within a short period of time, abnormal data reporting, or imbalanced operations.

[0065] When any of the above analysis conditions are met—that is, the behavior path fails to match the process node, or the fluctuation anomaly reaches the set warning criteria—the system generates an initial deviation label based on the corresponding anomaly type. This initial deviation label identifies an anomaly in the current behavior path, either in terms of structural logic or behavior frequency. It serves as a reminder that the behavior path needs to enter the next stage of further evaluation, including the calculation of behavior consistency scores and process integrity scores. This label serves as a key trigger in the entire evaluation methodology, ensuring that in-depth analysis is conducted only on data segments with potential issues, improving system efficiency and focus.

[0066] The logic for obtaining the behavior consistency score is as follows:

[0067] The path deviation value is calculated based on the behavior path graph. The path deviation value is obtained by counting the number of mismatched behaviors, key node jumps, and hysteresis operations in the interactive behavior path relative to the process logic, and dividing this total by the total number of operation steps defined in the standard path to obtain a normalized ratio reflecting the degree of structural deviation.

[0068] Fluctuation anomalies are calculated based on the operation frequency density model. The calculation of fluctuation anomalies is done by dividing user operation records and device status logs into multiple time periods of equal length. The change in the number of operation events within each time period is counted, and the absolute value of the difference in the number of operation events between adjacent time periods is processed and averaged. If a time period has a sudden change in the number of events, a local extreme value, or an abnormal change slope, a weight value greater than one is assigned to that time period to increase the impact of that period on the overall fluctuation. Specifically:

[0069] E i represents the number of events at the operation point in the i-th time period, E i-1 represents the number of events at the operation point in the i-1th time period, m is the total number of time periods, W is the uniform time period width, γ iIndicates the preset fluctuation weight factor. If the time period is a slope mutation point or an extreme point, the corresponding value is increased to >1, indicating an abnormal strengthening effect. If there is no slope mutation point or extreme point in the time period, the corresponding value = 1, indicating a normal effect. v Indicates fluctuation outliers.

[0070] The path deviation value and the fluctuation anomaly value are input into the rule-penalized scoring structure. The path deviation value is squared to increase the penalty intensity of structural anomalies, and the fluctuation anomaly value is squared to smooth out the high-frequency mutation anomalies, thereby generating a behavior consistency score. The score result is a continuous real number between zero and one. Specifically: D p is the path deviation value, reflecting the abnormal structure of process execution. α and β are adjustment parameters set by experience to balance the influence of structural deviation and fluctuation abnormality (such as α = 0.6, β = 0.4). CS represents the behavior consistency score, CS∈[0,1], the closer to 1, the more consistent the behavior.

[0071] The present invention introduces the idea of ​​edit distance in graph structure similarity analysis, and combines it with the interactive behavior structure characteristics of the behavior path graph to establish a path deviation value calculation mechanism, thereby realizing the quantification of the degree of structural deviation between the user's actual operation path and the standard process flow. The edit distance was originally used in image recognition and data structure comparison scenarios. The present invention migrates and optimizes it twice to measure whether there are structural anomalies such as node dislocation, key step skipping or response delay in the operation behavior, which significantly improves the ability to identify digital operation compliance and process consistency. In addition, the sensitivity of structural deviation is enhanced through square penalty processing, so that serious deviation behaviors obtain higher recognition weights in the scoring, thereby improving the abnormal identification ability of the scoring system.

[0072] At the same time, the present invention also integrates the sliding window variation rate analysis method in the field of signal processing, performs high-frequency fluctuation detection on the behavior density time series generated by the operation frequency density model, and identifies abnormal operation features such as local extreme values, slope mutations, and rhythm imbalances. This method is transformed into a fluctuation anomaly value calculation strategy based on the window density change rate and abnormal segment weight adjustment mechanism in the present invention, which improves the detection ability of abnormal behaviors such as centralized operations and false behavior reporting. By integrating structural deviations and fluctuation anomalies into behavioral consistency scores in the form of nonlinear penalty terms, a cross-domain method integration, quantitative accuracy, and strong explanatory power digital process behavior credibility determination mechanism is established, providing high-quality input for subsequent machine learning reasoning and credibility level output.

[0073] The logic for obtaining the process integrity score is:

[0074] The process chain breakage rate is calculated based on the temporal causal dependency model. The process chain breakage rate is calculated by counting the number of process dependencies that fail to be triggered in the interactive behavior path and dividing this number by the total number of dependencies in the predefined process flow. This reflects the proportion of process chains that are not activated in the interactive behavior.

[0075] Calculate the process conflict rate. The process conflict rate is calculated by identifying the number of process relationships that violate the dependency direction in the interactive behavior path and dividing this number by the total number of dependency relationships in the process flow. It reflects the proportion of behaviors in which the execution order is reversed or the dependency conditions are violated in actual execution.

[0076] Calculate the process coverage correction term. The process coverage correction term is formed by counting the number of process nodes successfully executed in the actual behavior path and dividing this number by the total number of nodes defined in the process to form a compensation factor to measure the completeness of the process execution;

[0077] The process chain breakage rate and process conflict rate are converted linearly and nonlinearly respectively, and the process coverage correction term is introduced to form the scoring function. The specific method is as follows: multiply the process chain breakage rate by the first weight parameter, multiply the square root of the process conflict rate by the second weight parameter, and use the sum of the above two terms to construct the basic scoring value in a one-minus form. The basic scoring value is then multiplied by the process coverage correction term to form the final process integrity score, which is a continuous real number between zero and one. Specifically:

[0078] F d Indicates the chain breakage rate (structural problem), F c represents the conflict rate (timing error), R cov is the coverage correction term used for positive compensation. λ1 and λ2 are parameter weights set empirically, such as one is 0.7 and the other is 0.3. The final process integrity score PIS∈[0,1] is obtained. The higher the value, the more complete and standardized the process.

[0079] In constructing a process integrity score, this paper innovatively incorporates the connectivity failure rate concept from the power system network fault transmission model and optimizes it into a process chain failure rate applicable to process flows. By counting the number of process dependencies that fail to activate in interactive behavior paths and combining it with the structural diagram of all dependencies in a standard process flow, a quantitative process chain breakage rate indicator is generated, which accurately reflects critical process path interruptions caused by missing or skipped behaviors. This approach shifts the logic of "connectivity breakage risk" in network systems to identifying "process structure gaps" in manufacturing scenarios, improving the accuracy of detecting structurally abnormal processes and is particularly suitable for detecting hidden chain breaks in complex, multi-stage manufacturing tasks. This paper also incorporates the control flow coverage concept from static code analysis in software engineering and transforms it into a process dependency coverage metric, serving as a positive correction mechanism for structural scores. By counting the proportion of successfully covered process nodes in the interactive behavior path to all process nodes, the company's actual operations can be intuitively reflected in their completeness in fulfilling the intended process. Introducing this coverage parameter into the scoring function not only prevents excessive score drops caused by minor process gaps, but also enhances the scenario adaptability and flexibility of the scoring. This method successfully migrates the control path execution concept from programming language analysis to the measurement system of industrial process node execution, forming a composite scoring mechanism that combines structural penalties with positive coverage adjustment, improving the accuracy and flexibility of process integrity judgment.

[0080] The machine learning model is implemented using a fuzzy logic controller. The fuzzy logic controller's inference method includes the following steps: A behavioral consistency score and a process integrity score are obtained and fed into the fuzzy logic controller as input variables. The behavioral consistency score represents the degree of consistency between user operation behavior and device response behavior in terms of structural path and frequency distribution, with a continuous value ranging from zero to one. The process integrity score represents the degree to which the interaction behavior adheres to the structure of the standard process flow, with a continuous value ranging from zero to one. Both scores are derived from the calculation results of the previous analysis model and have practical and quantifiable evaluation significance.

[0081] The two input variables are fuzzified based on multiple preset segmentation thresholds. Fuzzification refers to mapping continuous values ​​to fuzzy sets through membership functions and converting them into corresponding fuzzy linguistic variables. The system sets three levels of membership functions for each input variable: low level, medium level, and high level, which are used to indicate that the score value is in abnormal, critical, and normal states, respectively. For example, in the behavior consistency score, the low level may correspond to the [0.0–0.4] interval, the medium level corresponds to the [0.3–0.7] overlapping interval, and the high level corresponds to the [0.6–1.0] interval. The membership degree is divided by triangular or trapezoidal membership functions. This process enables the scoring results to have fuzzy description capabilities, which is conducive to subsequent rule judgment.

[0082] The output variable is set as the credibility level of information currently provided by the electromechanical industry, and is divided into three levels: high credibility, medium credibility, and low credibility, forming a fuzzy set of output variables. The credibility level comprehensively represents the credibility of the data in terms of behavioral structure and process logic, and is the core output criterion for evaluating the entire digitalization process.

[0083] Fuzzy rules are formulated based on the relationship between input variable grade combinations and output grades. These rules define the mapping between input grade combinations and output credibility levels. For example, if both the behavioral consistency score and the process integrity score are high, the output credibility level is high; if either score is medium, the output credibility level is medium; and if either score is low, the output credibility level is low. The system pre-sets all possible combinations and forms a complete fuzzy inference rule table.

[0084] The fuzzified input variables are subjected to fuzzy reasoning operations according to preset fuzzy rules. The reasoning operation matches the activated rule entries in the rule base based on the fuzzy membership of the input variables and generates the fuzzy membership results of the output variables at various credibility levels, forming a membership distribution.

[0085] The output variables are defuzzified based on the maximum membership principle. Defuzzification is the process of converting fuzzy membership distributions into definite values ​​or classification labels. Using the maximum membership method, the system selects the level with the highest membership among the trust levels as the trust level output corresponding to the information provided by the current electromechanical industry.

[0086] For example, a batch of digitized behavioral data from a certain enterprise had a behavioral consistency score of 0.78 and a process integrity score of 0.81. Based on the pre-defined membership function, the system determined that both scores belonged to the "high" fuzzy set. The "high + high → high confidence" rule in the fuzzy rule base was activated, and after reasoning, the output confidence level reached 0.88 on the "high confidence" membership scale, significantly higher than other levels. After performing the defuzzification process, the system ultimately determined that the data was "highly trustworthy" and allowed it to proceed to the digitization process evaluation phase.

[0087] High confidence, medium confidence and low confidence are the three levels of confidence of the system's final output. Among them, high confidence means that the information provided by the electromechanical industry meets the preset confidence requirements, and the information provided by the electromechanical industry is suitable for the final digital process evaluation, and the digital process evaluation is carried out according to the normal process; medium confidence and low confidence mean that the information has problems in structural consistency, process integrity or operation density stability, and does not meet the preset confidence requirements. The system needs to propose information optimization strategies based on their corresponding levels and feedback to the electromechanical industry.

[0088] The information optimization strategy corresponding to medium credibility is: locating local deviation areas and prompting correction suggestions. Medium credibility level means that any one of the behavioral consistency score and process integrity score is in a medium-level state, indicating that the data has certain slight structural deviations, abnormal path sequence or operation frequency fluctuations, but has not reached the severe abnormality threshold. The system will combine the initial deviation label with the output results of the scoring model to locate the specific behavioral paragraphs or time segments where the deviation occurs, mark the user behavior graph nodes or process dependency nodes involved, and generate correction prompt information, including but not limited to: prompting users to supplement missing operation records, correct incorrect time sequence, check process logic configuration, etc. This type of strategy does not require data re-collection, but guides enterprises to conduct data verification and lightweight corrections to improve the data credibility level to meet the usable standard.

[0089] The information optimization strategy corresponding to low credibility is: refuse to adopt the current data and recommend re-collection or process reconstruction. Low credibility level means that at least one of the behavioral consistency score and the process integrity score is at a low level, indicating that the data segment has serious anomalies in the path structure, process logic or operation behavior, which may include serious disorder of operation sequence, missing key nodes, and extremely abnormal operation density. While identifying the problem characteristics, the system will generate a feedback report, clearly indicating the scope of unreliable data segments, corresponding problem categories and abnormal dimensions, such as "the device did not respond but the process continued to execute" and "the abnormal density is concentrated in short-term operations of a single user". The feedback results will be accompanied by collection rule suggestions, such as extending the operation time period, improving the accuracy of log collection, resetting the trigger conditions of key nodes, etc., prompting enterprises to make structural corrections to the collection logic, system configuration or process modeling methods at the source level to ensure that the re-collected data meets the credibility standards.

[0090] Example 1 (Medium Credibility): In an assembly line operation record uploaded by a company, the user behavior path generally complies with process requirements. However, there is a sequence advance between the "positioning" and "screwing" steps, causing the process integrity score to drop to 0.68, resulting in a medium rating. The system feedback prompts: "The process sequence of records 9-12 is suspected to be abnormal. It is recommended to verify the operation time and correct the configuration."

[0091] Example 2 (Low Credibility): A workshop uploaded 80 "start" commands within one minute. The operation frequency density model identified severe abnormal fluctuations, a behavioral consistency score below 0.35, and a lack of "sensor trigger response" in the behavioral path. The system generated an optimization suggestion: "The credibility of the current data segment is too low. Please check the collection mechanism for centralized backfilling or script injection. We recommend re-collecting and encrypting the operation entry."

[0092] In the present invention, the behavioral consistency score and process integrity score are evaluated based on a subset of samples for which initial deviation labels have been generated. The system does not initially calculate a score for each data segment. Instead, it uses a pre-process—"analysis matching and fluctuation detection"—to determine whether anomalies exist before entering the evaluation phase. This pre-process itself includes: item-by-item comparison of the behavioral path graph with process nodes to determine whether operation-response pairs match process dependencies; and mutation detection in the operation frequency density model to locate density anomalies by time period. Therefore, during the generation of the "initial deviation label," the system already records which behavioral path segment (i.e., which user behavior graph nodes and device response graph nodes are included) and within which time period triggered the "structural deviation" or "fluctuation anomaly." This information is incorporated into the generated deviation label as raw metadata regarding the location and type of the anomaly. It is then used to generate feedback recommendations when the anomaly is determined to be "medium confidence" or "low confidence."

[0093] Throughout the entire invention process, whether or not an initial deviation label is generated is the trigger condition for entering the evaluation phase. If a certain data segment, under the three-type model analysis, does not reveal the following conditions: behavioral paths and process nodes are in reverse order, step skipping, or response lag; operation density fluctuates dramatically; or any structural or behavioral deviations exceed the threshold, then the data segment will not be marked as an initial deviation sample, indicating that its structure and behavior are normal. At this point: the system will no longer score or infer this segment of data because there are no anomalies; the data will be directly classified as "high confidence level" and directly enter the final digital process evaluation based on the rule of "high confidence meets preset confidence requirements."

[0094] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters in the formulas are set by technicians in this field according to actual conditions.

[0095] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0096] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0097] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0098] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A method for evaluating the digitalization process of the electromechanical industry based on data analysis, characterized in that: The following steps are involved: User operation records, equipment status logs, and process execution data are extracted from information provided by the electromechanical industry. Through structural analysis, a behavior path map, a temporal causal dependency model, and an operation frequency density model are established. The behavior path map describes the operation path between users and equipment, the temporal causal dependency model reflects the process execution sequence and logical constraints, and the operation frequency density model describes the behavior fluctuation characteristics within a unit of time. Perform analysis and matching to determine whether to generate an initial deviation label. If an initial deviation label has been generated, perform further evaluation: Calculate a behavioral consistency score based on the degree of path deviation in the behavioral path map and the degree of fluctuation anomaly in the operation frequency density model. Calculate a process integrity score based on the logical chain breakpoints and dependency conflict degree in the temporal causal dependency model. The behavior consistency score and process integrity score are used as inputs to a machine learning model pre-trained with historical data from the electromechanical industry to perform risk reasoning and output the corresponding trust level of the information provided by the current electromechanical industry. The following judgments are made based on the credibility level: The information provided by the electromechanical industry meets the preset trust requirements and is suitable for the final digital process evaluation, and the digital process evaluation is carried out according to the normal process; The information provided by the electromechanical industry does not meet the preset trust requirements and is not suitable for the final digital process evaluation. Based on the trust level that does not meet the preset trust requirements, a corresponding information optimization strategy is proposed and fed back to the electromechanical industry.

2. The method for evaluating the digitalization process of the electromechanical industry based on data analysis according to claim 1 is characterized in that: The construction of the behavior path map includes the following processes: Sort the user operation records and device status logs in the information provided by the electromechanical industry in chronological order, extract the operation time, operation subject, operation action and device response information, and generate user behavior graph nodes and device response graph nodes; Directed graph edges are constructed between graph nodes in the form of action-response pairs to form a continuous interaction path between the user and the device. The behavior path graph is a set of behavior sequence graph structures with sequential constraints. Each path is initiated by a specific user and connects multiple device response graph nodes in series according to time logic.

3. The method for evaluating the digitalization process of the electromechanical industry based on data analysis according to claim 2 is characterized in that: The construction of a temporal causal dependency model includes the following steps: Pre-process the process execution data in the information provided by the electromechanical industry, treat each process as a process dependency node, extract the start time, end time, and task trigger conditions of each process, and construct a process dependency table based on the execution sequence in the process flow; A directed acyclic graph structure is generated based on the dependency table. The direction of the edge in the graph represents the process dependency sequence. Each process dependency node contains a time stamp and a trigger condition field to determine whether the process sequence is violated.

4. The method for evaluating the digitalization process of the electromechanical industry based on data analysis according to claim 3 is characterized in that: The construction of the operation frequency density model includes the following processes: User operation records and device status logs are divided into time periods of equal length. The total number of corresponding operation events in each time period is calculated to generate a time period operation point sequence. Each time period operation point records the number of operation events and operation category distribution information within the time period. An operation density curve is constructed based on the changing trend of the number of operation events, and local extreme values, slope mutation points, and change rate information are extracted from the curve to jointly reflect the intensity of behavioral fluctuations.

5. The method for evaluating the digitalization process of the electromechanical industry based on data analysis according to claim 4 is characterized in that: Performing analysis and matching and deciding whether to generate an initial deviation label refers to: The operation paths in the behavior path map are matched with the process nodes in the temporal causal dependency model to determine whether the interactive behavior has order reversal, step skipping, or response lag relative to the process logic. At the same time, based on the sharp fluctuations in the number of operation events in the corresponding time period in the operation frequency density model, it is determined whether there is a density anomaly. If the match fails or the fluctuation anomaly reaches the preset anomaly threshold, an initial deviation label is generated based on the corresponding anomaly type.

6. The method for evaluating the digitalization process of the electromechanical industry based on data analysis according to claim 5 is characterized in that: The logic for obtaining the behavior consistency score is as follows: The path deviation value is calculated based on the behavior path graph. The path deviation value is obtained by counting the number of mismatched behaviors, key node jumps, and hysteresis operations in the interactive behavior path relative to the process logic, and dividing this total by the total number of operation steps defined in the standard path to obtain a normalized ratio reflecting the degree of structural deviation. Fluctuation anomalies are calculated based on the operation frequency density model. This is done by dividing user operation records and device status logs into multiple time periods of equal length. The magnitude of the change in the number of operation events within each time period is then counted. The absolute value of the difference in the number of operation events between adjacent time periods is processed and averaged. If a time period experiences a sudden change in the number of events, a local extreme value, or an abnormal slope of change, a weight greater than one is assigned to that time period to increase its impact on the overall fluctuation. The path deviation value and the fluctuation anomaly value are input into the rule-penalized scoring structure together. The path deviation value is squared to enhance the penalty intensity of structural anomalies, and the fluctuation anomaly value is squared to smooth out the high-frequency mutation anomalies, thereby generating a behavioral consistency score, which is a continuous real number between zero and one.

7. The method for evaluating the digitalization process of the electromechanical industry based on data analysis according to claim 6 is characterized in that: The logic for obtaining the process integrity score is: The process chain breakage rate is calculated based on the temporal causal dependency model. The process chain breakage rate is calculated by counting the number of process dependencies that fail to be triggered in the interactive behavior path and dividing this number by the total number of dependencies in the predefined process flow. This reflects the proportion of process chains that are not activated in the interactive behavior. Calculate the process conflict rate. The process conflict rate is calculated by identifying the number of process relationships that violate the dependency direction in the interactive behavior path and dividing this number by the total number of dependency relationships in the process flow. It reflects the proportion of behaviors in which the execution order is reversed or the dependency conditions are violated in actual execution. Calculate the process coverage correction term. The process coverage correction term is formed by counting the number of process nodes successfully executed in the actual behavior path and dividing this number by the total number of nodes defined in the process to form a compensation factor to measure the completeness of the process execution; The process chain breakage rate and process conflict rate are converted linearly and nonlinearly respectively, and the process coverage correction term is introduced to form the scoring function. The specific method is: multiply the process chain breakage rate by the first weight parameter, multiply the square root of the process conflict rate by the second weight parameter, and use the sum of the above two items to construct the basic scoring value in the form of one minus one, and then multiply it by the process coverage correction term to form the final process integrity score. The scoring value is a continuous real number between zero and one.

8. The method for evaluating the digitalization process of the electromechanical industry based on data analysis according to claim 7 is characterized in that: The machine learning model is implemented using a fuzzy logic controller. The reasoning method of the fuzzy logic controller includes the following steps: Obtain the behavior consistency score and process integrity score, and input them into the fuzzy logic device as input variables; then perform fuzzification on the two input variables according to multiple preset segmentation thresholds, and divide the input variables into three fuzzy sets of low level, medium level and high level; then set the output variable as the credibility level of the information provided by the current electromechanical industry, and divide the output variable into three levels of high credibility, medium credibility and low credibility, forming a fuzzy set of output variables; formulate fuzzy rules based on the relationship between the input variable level combination and the output level, perform fuzzy reasoning operations on the fuzzified input variables according to the preset fuzzy rules, and obtain the fuzzy membership result corresponding to the output credibility level, and defuzzify the output variable based on the maximum membership principle, and finally output the credibility level corresponding to the information provided by the current electromechanical industry.

Images