Intelligent monitoring method and system for numerical control equipment based on machining progress
By constructing a causal directed graph and a baseline correlation model, the problem of correlation between progress anomalies and equipment status, process execution, and alarm events in CNC equipment was solved, realizing the location of the root cause of the anomaly and improving the stability of production.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- DONGGUAN ZHIYUAN CNC EQUIP MFG CO LTD
- Filing Date
- 2026-05-26
- Publication Date
- 2026-06-23
AI Technical Summary
Existing CNC equipment monitoring and control and data acquisition systems struggle to correlate issues such as machining cycle time fluctuations, program segment delays, and cumulative progress lags with equipment status, process execution, and alarm events in a unified manner. This leads to a disconnect between anomaly identification and root cause localization, relying on manual experience for troubleshooting, which is insufficient to support stable production and refined scheduling.
By acquiring progress data, equipment status data, process execution data, and alarm codes of CNC equipment, a causal directed graph is constructed, a baseline correlation model is established, and probabilistic reasoning is performed using deviation vectors to determine the root cause nodes and propagation paths of anomalies.
It achieves a unified quantitative representation of the cycle time status of the processing, which can transform progress anomalies into comparable baseline deviation results, locate the root cause of the anomaly, support maintenance troubleshooting, process review and production adjustment, and improve production stability.
Smart Images
Figure CN122261019A_ABST
Abstract
Description
Technical Field
[0001] This application belongs to the field of monitoring and control technology, specifically to a method and system for intelligent monitoring of CNC equipment based on processing progress. Background Technology
[0002] CNC equipment is a core component of discrete manufacturing, and its operational status directly impacts machining cycle time, product consistency, equipment utilization, and production organization stability. Existing monitoring, control, and data acquisition systems typically collect spindle load, feed status, program execution information, and alarm information, displaying or alerting to abnormal states. However, most methods remain at the level of single-parameter limit judgment, alarm code reporting, or post-event traceability. For issues such as machining cycle time fluctuations, program segment delays, and cumulative progress lags, existing monitoring, control, and data acquisition systems often struggle to unify the correlation between process execution changes, equipment status changes, progress anomalies, and alarm events. This leads to a disconnect between anomaly identification and root cause localization, resulting in significant reliance on manual experience for on-site troubleshooting, which is insufficient for timely support of stable production, predictive maintenance, and refined scheduling. Summary of the Invention
[0003] To address the above issues, this application provides a method and system for intelligent monitoring of CNC equipment based on machining progress, which at least solves the problem of how to correlate abnormal progress with equipment status, process execution, and alarm events during CNC equipment machining and locate the root cause of the abnormality.
[0004] To achieve the above objectives, the technical solution adopted in this application is as follows: In a first aspect, this application provides a method for intelligent monitoring of CNC equipment based on machining progress, the method comprising: Acquire the progress data, equipment status data, process execution data and alarm codes of the CNC equipment, and determine the progress reference benchmark corresponding to the program segment based on historical normal processing data. The progress data includes the actual execution time and cumulative completion progress of the program segment, and the progress reference benchmark includes the reference execution time and reference completion progress of the program segment. The progress data, equipment status data, process execution data, and alarm codes are time-aligned and correlated according to program segment numbers to obtain a fused dataset; A causal directed graph including process execution nodes, equipment status nodes, processing progress nodes, and alarm code nodes is constructed based on the fused dataset. The edge directions in the causal directed graph are constrained based on the processing sequence constraints, and a baseline association model is established. When a deviation from the progress reference baseline is detected or an alarm code is generated, the deviation vector is determined based on the current observation value and the predicted value corresponding to the baseline correlation model. Based on the deviation vector, probabilistic reasoning is performed on the causal directed graph to determine and output the abnormal root cause node and abnormal propagation path.
[0005] Secondly, this application provides an intelligent monitoring system for CNC equipment based on machining progress, used to implement an intelligent monitoring method for CNC equipment based on machining progress. The system includes: The data acquisition module is used to acquire the progress data, equipment status data, process execution data and alarm codes of the CNC equipment, and determine the progress reference benchmark corresponding to the program segment based on historical normal machining data. The progress data includes the actual execution time and cumulative completion progress of the program segment, and the progress reference benchmark includes the reference execution time and reference completion progress of the program segment. The data fusion module is used to align progress data, equipment status data, process execution data, and alarm codes by time, and associate them according to program segment numbers to obtain a fused dataset. The model building module is used to construct a causal directed graph based on the fused dataset, including process execution nodes, equipment status nodes, processing progress nodes, and alarm code nodes, and to establish a baseline association model by limiting the edge directions in the causal directed graph based on the processing sequence constraints. The inference output module is used to determine the deviation vector based on the current observation value and the predicted value corresponding to the benchmark correlation model when the processing progress deviates from the progress reference benchmark or an alarm code is detected. Based on the deviation vector, probabilistic inference is performed on the causal directed graph to determine and output the abnormal root cause node and abnormal propagation path.
[0006] Compared with existing technologies, the advantages and beneficial effects of this application are as follows: By employing multi-source processing data joint acquisition and program segment reference benchmark construction techniques, a unified quantitative representation of the processing cycle status is achieved, transforming progress anomalies from fuzzy empirical judgments into comparable benchmark deviation results. Time alignment and program segment association techniques enable the homogeneous organization of progress data, equipment status data, process execution data, and alarm codes, facilitating subsequent analysis within the same temporal framework. The introduction of causal directed graph construction techniques with processing sequence constraints enables modeling of the transmission relationships between process changes, status changes, progress anomalies, and alarm events, avoiding isolated judgments based on single-point signals. The establishment of a benchmark association model based on historical normal processing data allows for the extraction of differences between the current observed state and normal operating conditions, facilitating the formation of deviation descriptions with direction, amplitude, and persistence. Deviation vector-driven probabilistic inference techniques enable the location output of anomaly root cause nodes and anomaly propagation paths, not only indicating anomaly occurrences but also supporting maintenance troubleshooting, process review, and production adjustments. Attached Figure Description
[0007] Figure 1 This is a flowchart illustrating the method described in this application; Figure 2 This is a block diagram of the module composition of the system in this application. Detailed Implementation
[0008] To enable those skilled in the art to better understand the technical solution, the present application will be described in detail below with reference to the embodiments. The description in this section is only exemplary and explanatory, and should not be used to limit the scope of protection of the present application in any way.
[0009] Processing progress refers to the actual advancement status of CNC equipment during the execution of a processing task, relative to a predetermined process path, program segment sequence, and expected cycle time. It not only reflects the current processing stage of the workpiece but also the stability of the execution transitions between program segments, the timeliness of equipment response, and the continuity of the process. In actual processing, processing progress is not a single time quantity but a comprehensive process state closely related to the actual execution time of program segments, cumulative completion rate, process switching rhythm, and the insertion of abnormal events. Therefore, combining continuous perception and structured representation of processing progress with a monitoring and control system and a data acquisition system can provide direct evidence for identifying cycle time deviations, delay accumulation, and anomaly propagation during the processing. Based on this, this application focuses on processing progress—a key process quantity that connects process execution, equipment status, and abnormal behavior—and constructs an intelligent monitoring technology solution for CNC equipment that integrates monitoring and control and data acquisition systems.
[0010] like Figure 1 As shown, a method for intelligent monitoring of CNC equipment based on machining progress is described, the method comprising: Acquire the progress data, equipment status data, process execution data and alarm codes of the CNC equipment, and determine the progress reference benchmark corresponding to the program segment based on historical normal processing data. The progress data includes the actual execution time and cumulative completion progress of the program segment, and the progress reference benchmark includes the reference execution time and reference completion progress of the program segment. In one embodiment, after the CNC equipment enters a machining task, it first acquires the progress data, equipment status data, process execution data, and alarm codes corresponding to the current machining task. The progress data characterizes the advancement of the program segment during the actual machining process, including at least the actual execution time and cumulative completion progress of the program segment. The equipment status data characterizes the current operating status of the machine tool, the process execution data characterizes the current program call and process settings, and the alarm codes characterize abnormal events that occur during the machining process. After data acquisition, the current machining task is matched with historical normal machining data, and historical samples with consistent equipment model, machining process, workpiece type, and normal machining results are selected. Then, based on the execution records of each program segment in the historical samples, a progress reference benchmark corresponding to the program segment is determined. The progress reference benchmark includes at least the reference execution time and reference completion progress of the program segment. After obtaining the progress reference benchmark, the progress data, along with the equipment status data, process execution data, and alarm codes, are sent to the next stage for subsequent time alignment and program segment association processing.
[0011] The progress data includes the actual execution time of the program segment and the cumulative completion progress. The progress reference benchmark includes the reference execution time of the program segment and the reference completion progress. The progress reference benchmark is determined by the historical normal processing data corresponding to the current processing task.
[0012] In one embodiment, to ensure that the progress reference benchmark accurately reflects the normal cycle time of the current machining task, this embodiment further adds task matching rules, sample screening rules, and benchmark update rules during the data acquisition and benchmark generation stages. Specifically, after the CNC equipment starts executing the machining program, the data acquisition unit first reads the program segment number, program start time, program segment switching time, feed execution status, spindle running status, and alarm event records of the current machining task from the CNC system, servo drive unit, and machine tool control interface. The actual execution time of a program segment is obtained from the time difference between the switching times of two adjacent program segments; when a program segment is the currently executing program segment, the actual execution time of the program segment is obtained from the time difference between the current acquisition time and the start time of that program segment. The cumulative progress is used to characterize the degree of progress of the current machining task in the overall machining process, and can be determined according to the proportion of the number of completed program segments to the total number of program segments, or according to the proportion of the standard working time of the completed program segments to the total standard working time of the current machining task. If the machining program includes an idle run segment, a tool change segment, or a manual confirmation segment, the above program segments are first identified based on the program segment attribute tags. Then, program segments that do not participate in the effective machining cycle count are removed or recorded separately to avoid affecting the subsequent benchmark generation results.
[0013] In the process of screening historical normal processing data, the first layer of screening criteria is the equipment number, workpiece category, process version, and tool configuration corresponding to the current processing task, which are used to retrieve candidate samples from the historical database. The second layer of screening criteria is the quality inspection results, downtime records, and alarm records after processing, which remove samples with out-of-tolerance, downtime for maintenance, abnormal interruptions, or critical alarms, retaining samples that successfully completed the processing task as historical normal processing data. Successfully completing a processing task here means that the processing flow ends according to the predetermined procedure, the product quality meets acceptance requirements, and no abnormal events occur during processing that would cause task interruption. To avoid distortion of the baseline due to an insufficient number of historical samples or an excessively long sample time span, this embodiment can also set a minimum sample number and a valid sample time range. The minimum sample number can be set to five to twenty batches of samples from the most recent continuous production, and the valid sample time range can be set to samples from the most recent month, the most recent quarter, or the most recent process version cycle. When the number of historical samples meeting the first and second layer screening criteria is insufficient, the tool configuration consistency condition can be relaxed, but the equipment number, workpiece category, and process version consistency condition is retained to ensure that the progress reference baseline still has a direct correspondence with the current processing task.
[0014] After sample screening, historical normal processing data is segmented and statistically analyzed according to program segment number. For each program segment, the execution time of the program segment in each historical sample is extracted, and discrete values that deviate significantly from the normal range are removed to determine the reference execution time of the program segment. Discrete values can be eliminated according to a pre-set upper limit of deviation; for example, data exceeding a certain proportion of the median value in the same batch of samples are not included in the statistics. The reference completion progress is determined based on the sequential position of each program segment in the entire processing task and its proportion of standard working time, so that each program segment has a comparable completion progress node at the end. In this way, during actual processing, once the current program segment number and the current execution time are read, the reference execution time and reference completion progress corresponding to that program segment can be obtained synchronously. To adapt to changes in equipment status, tool changes, and process fine-tuning, this embodiment can also write the batch data into the historical database according to the same screening rules after each batch of processing tasks is completed and accepted, and the progress reference benchmark of the corresponding program segment is updated on a rolling basis. During the rolling update, recent samples are used first to make the benchmark closer to the current equipment and process status. Using the above methods, a progress reference benchmark that is clearly derived, updates are controllable, and can be directly used for subsequent anomaly monitoring can be formed without introducing complex mathematical models.
[0015] The progress data, equipment status data, process execution data, and alarm codes are time-aligned and correlated according to program segment numbers to obtain a fused dataset; In one embodiment, after obtaining progress data, equipment status data, process execution data, and alarm codes, time alignment and program segment association processing are performed on various types of data. Specifically, a unified time base is first selected, and data with different sampling periods and recording formats are converted to a unified time axis. Then, a correspondence between the data and the current processing program segment is established based on the program segment number. After time unification is completed, the associated data is organized according to the equipment number, processing task number, program segment number, and time window to form a fused dataset. The fused dataset is used to simultaneously represent the progress status, equipment operating status, process execution status, and abnormal event status of the current processing stage within the same data unit, so that it can be directly called upon for subsequent causal relationship construction and baseline association modeling.
[0016] Time alignment and program segment association include: timestamping and resampling progress data, equipment status data, process execution data, and alarm codes according to a unified time base; establishing the correspondence between the corrected data and program segments based on the program segment number; and organizing the associated data according to the equipment number, processing task number, program segment number, and time window to obtain a fused dataset.
[0017] In one embodiment, in order to enable the time alignment and program segment association results to be directly used for subsequent anomaly monitoring and processing, the determination method of the unified time base, resampling rules, program segment mapping rules, and organization structure of the fused dataset are further defined based on the above processing.
[0018] Specifically, a unified time reference can be established using either the CNC system's control clock or the synchronous clock in an industrial fieldbus. When both the equipment and the host control system have independent clocks, the timestamp corresponding to the most recent successful communication from each data source is first read to establish a clock offset. Then, timestamp correction is performed based on the clock offset, ensuring that progress data, equipment status data, process execution data, and alarm codes are all mapped to the same reference time axis. Progress data is typically recorded according to program segment switching events and processing progress refresh events; equipment status data is typically collected at a fixed sampling period; process execution data is typically recorded according to program call or parameter change events; and alarm codes are typically recorded according to the event trigger time.
[0019] For the aforementioned heterogeneous recording methods, the time window can be set to a fixed-length window or a program segment adaptive window. A fixed-length window is suitable for processing scenarios with stable sampling periods, such as using a statistical window of one to five seconds. A program segment adaptive window is suitable for processing scenarios with large differences in program segment duration, i.e., using the start and end times of the current program segment as the window boundary. During resampling, continuous data is written to the corresponding window using interpolation of nearby sampling points or the window average, while event-type data is written to the corresponding window using event endpoint retention, avoiding the expansion of alarm events into unrealistic continuous states. After time unification is completed, a correspondence between the corrected data and the program segment is established based on the program segment number. If a program segment switch occurs within the same time window, the data within the window is split and written to the preceding and following program segments, using the switch time as the boundary. If there are interruption events such as pause, tool change, or reset, a status flag is written separately, and the interruption reason is retained in the program segment association results, avoiding misjudging data during interruptions as normal processing data.
[0020] The fused dataset can be organized using either a row-based data table structure or a key-value mapping structure. Regardless of the structure, each data unit must contain at least the following fields: equipment number, processing task number, program segment number, start and end times of the time window, progress data field, equipment status data field, process execution data field, and alarm code field. This results in data units that reflect the complete processing status of a program segment within a specific time window and can be seamlessly integrated with preceding and following time windows, thus providing a continuous, homogeneous, and comparable data foundation for subsequent causal directed graph construction.
[0021] A causal directed graph including process execution nodes, equipment status nodes, processing progress nodes, and alarm code nodes is constructed based on the fused dataset. The edge directions in the causal directed graph are constrained based on the processing sequence constraints, and a baseline association model is established. In one embodiment, after obtaining the fused dataset, a causal directed graph is first constructed based on the sequential relationships in the processing flow. Specifically, four types of variables—process execution, equipment status, processing progress, and alarm codes—are extracted from the fused dataset, and corresponding nodes are formed for each. Then, the edge directions are determined according to the order in which program calls precede equipment responses, equipment responses precede progress changes, and alarm occurrences, resulting in a graph structure reflecting the transmission relationships in the processing process. After completing the causal directed graph construction, historical normal processing data is retrieved, and the upstream inputs of each node and the target output of the current node are determined according to the directed connections between nodes. Predictive relationships between each node are established, and all predictive relationships are combined to form a baseline correlation model. The baseline correlation model and the causal directed graph together serve as the basis for subsequent anomaly deviation analysis and root cause reasoning.
[0022] A causal directed graph includes process execution nodes, equipment status nodes, processing progress nodes, and alarm code nodes. Processing sequence constraints are used to limit the direction of edges in the causal directed graph, so that process execution nodes point to equipment status nodes, and equipment status nodes point to processing progress nodes or alarm code nodes.
[0023] In one embodiment, in order to enable the causal directed graph to truly reflect the state transmission relationship in the CNC machining process, this embodiment further defines the method of determining the node composition and edge direction based on the foregoing content.
[0024] Specifically, the fields in the fused dataset are first categorized. Program segment number, set feed rate, set spindle speed, tool number, and process switching flag are assigned to the process execution node. Spindle load, feed axis current, vibration amplitude, critical component temperature, and position tracking deviation are assigned to the equipment status node. The actual execution time of the program segment, program segment completion flag, and cumulative completion progress are assigned to the machining progress node. Alarm codes, alarm occurrence time, and alarm duration are assigned to the alarm code node. After node categorization, edge directions are established according to the temporal relationship in the machining process. The process execution node represents machining instructions and process settings; its changes precede equipment response, therefore it points to the equipment status node. The equipment status node represents the machine tool's operating results after executing process instructions; its changes further affect machining cycle time and abnormal events, therefore it points to either the machining progress node or the alarm code node.
[0025] To avoid creating unrealistic reverse relationships, edges pointing from processing progress nodes to process execution nodes and from alarm code nodes to equipment status nodes are not established during the graph construction process. If multiple equipment status fields exist within the same program segment, representative values for each field within a time window are first calculated, and then these representative values are written to the corresponding nodes. If multiple alarm codes appear within the same time window, different alarm code nodes are generated and connected to the equipment status nodes within the same time window. After generating the nodes and edges, a graph structure that expands with the program segment and time window is obtained. This graph structure preserves both the sequential order of the processing and the path of equipment status propagation to progress anomalies and alarm events, providing a direct data foundation for subsequent screening of anomaly propagation paths.
[0026] Establishing a baseline correlation model includes: determining the upstream nodes of each node based on the causal directed graph; using the data corresponding to the upstream nodes of each node as input and the values of each node in historical normal processing data as output, establishing the prediction relationship of each node respectively; and establishing a baseline correlation model based on the prediction relationship of each node.
[0027] In one embodiment, in order to enable the benchmark correlation model to stably represent the dependencies between various nodes under normal operating conditions, this embodiment further defines the selection method of historical normal processing data, the determination method of upstream nodes, and the establishment method of node prediction relationships based on the foregoing content.
[0028] Specifically, the processing records corresponding to the current processing task are first retrieved from the historical database. Historical normal processing data is then filtered based on criteria such as consistent equipment model, consistent workpiece category, consistent process version, and acceptable processing results. Historical records with downtime, critical alarms, or quality defects are excluded from the current modeling. After sample filtering, the upstream nodes of each node are determined one by one according to the edge directions in the causal directed graph. The process execution node serves as the starting node; it may not have an upstream node set, or the program segment identifier and process settings may be used as the node's own input. The equipment status node takes the data corresponding to the upstream process execution node as input and the equipment status value within the current time window as output. The processing progress node and alarm code node take the data corresponding to the upstream equipment status node as input and the progress value or alarm status within the current time window as output.
[0029] Subsequently, prediction relationships are established for each type of node. These relationships can be implemented using regression, lookup table mapping, or trained nonlinear mapping units. Regardless of the form, the input remains the data from the corresponding upstream node, and the output remains the value of the current node in historical normal processing data. After establishment, the prediction relationships of each node are combined according to the connection order in the causal directed graph to form a baseline correlation model covering process execution, equipment status, processing progress, and alarm codes. To ensure model usability, a consistency check is performed on the prediction results of each node before combination. If the deviation of a node's prediction relationship in recent samples consistently exceeds a preset range, the most recently completed normal samples are retrieved for updating. This preset range can be determined based on the normal fluctuation range of the corresponding node in historical normal processing data, ensuring that the baseline correlation model reflects normal processing patterns without being distorted by individual abnormal samples. Through this method, a baseline correlation model directly corresponding to the current processing task and usable for subsequent deviation vector extraction can be obtained.
[0030] When a deviation from the progress reference baseline is detected or an alarm code is generated, the deviation vector is determined based on the current observation value and the predicted value corresponding to the baseline correlation model. Based on the deviation vector, probabilistic reasoning is performed on the causal directed graph to determine and output the abnormal root cause node and abnormal propagation path.
[0031] In one embodiment, after constructing the fused dataset, the causal directed graph, and the baseline correlation model, the monitoring system continuously receives data corresponding to the current moment and performs anomaly trigger judgment based on the progress reference baseline and alarm code. When the actual execution time of the program segment exceeds the reference execution time of the program segment, or the cumulative completion progress is lower than the reference completion progress, or an alarm code occurs at the current moment, the current observation value is extracted from the fused dataset, and the corresponding predicted value is obtained by combining it with the baseline correlation model. Then, a deviation vector is generated based on the difference direction, difference magnitude, and difference duration. Subsequently, using the deviation vector as anomaly evidence, the anomaly probability of each node and its contribution to symptom nodes are calculated on the causal directed graph to determine the root cause node and the anomaly propagation path. The location results, handling suggestions, and event records are then output to the monitoring interface and event database.
[0032] If the processing progress is detected to deviate from the progress reference baseline, including at least one of the following: the actual execution time of the detected program segment is longer than the reference execution time of the detected program segment; or the cumulative completion progress is lower than the reference completion progress.
[0033] In one embodiment, this embodiment further defines the triggering rules for the deviation of the processing progress from the progress reference benchmark, relative to the foregoing, in order to clarify the boundary conditions for the initiation of anomaly analysis and avoid misjudging normal fluctuations of the program as anomalies.
[0034] Specifically, after each sampling cycle, the monitoring system first reads the current program segment number, the start time of the current program segment, the current sampling time, and the cumulative completion progress. Then, it reads the reference execution time and reference completion progress of the program segment corresponding to the current program segment from the progress reference baseline. The actual execution time of the program segment is obtained by subtracting the start time of the current program segment from the current sampling time; when the program segment has ended, the final value can also be obtained by subtracting the start time of the program segment from the end time of the program segment. The cumulative completion progress is used to characterize the progress ratio of the current processing task in the entire processing flow. It can be expressed as the ratio of the number of completed program segments to the total number of program segments, or as the ratio of the completed standard working hours to the total standard working hours.
[0035] To adapt to different process cycles and equipment states, the progress deviation judgment does not directly use a single instantaneous comparison, but rather a combined rule of threshold and duration. For the actual execution time of a program segment, a duration deviation threshold can be set. Only when the actual execution time exceeds the reference execution time of the program segment and the exceedance reaches the duration deviation threshold is the current program segment considered to have a duration anomaly. The duration deviation threshold can be determined based on the normal fluctuation range of the execution time of the corresponding program segment in historical normal processing data. For example, it can be the deviation corresponding to the upper limit of the duration fluctuation in the most recent normal sample, or a certain percentage of the reference execution time.
[0036] For cumulative progress, a progress lag threshold can be set. A progress anomaly is only identified when the cumulative progress falls below the reference progress and the lag remains continuously for a preset time. This is because factors such as program switching, cache refreshing, and temporary load fluctuations may cause slight deviations in a short period, but do not necessarily indicate a true anomaly. Alarm code triggering is relatively straightforward. When an alarm code appears within the current time window, anomaly analysis can be initiated immediately. If the alarm code is a non-fault code such as a tool change reminder, maintenance reminder, or manual confirmation prompt, it is first filtered according to the alarm code classification table, and anomaly analysis is initiated only for alarm codes directly related to machining anomalies, such as fault alarms, overload alarms, servo alarms, and temperature alarms. If both progress deviation and alarm code triggering are satisfied within the same time window, both trigger results are written into the anomaly analysis task as the input basis for subsequent deviation vector generation and symptom node determination.
[0037] Determining the bias vector includes: extracting the current observation from the fusion dataset corresponding to the current time; determining the predicted value corresponding to the current observation based on the benchmark correlation model; and generating the bias vector based on the direction, magnitude, and duration of the difference between the current observation and the predicted value.
[0038] In one embodiment, this embodiment further defines the method of generating the deviation vector, relative to the foregoing, to convert the difference between the current processing state and the normal working condition into unified anomalous evidence that can be used for causal reasoning.
[0039] Specifically, after anomaly analysis is initiated, the monitoring system first extracts the current observations from the fused dataset corresponding to the current moment. The current observations include progress data, equipment status data, process execution data, and alarm codes corresponding to the current program segment. The progress data includes at least the actual execution time and cumulative progress of the program segment. The equipment status data may include spindle load, feed axis current, critical component temperature, vibration amplitude, and position tracking deviation. The process execution data may include the program segment number, set feed rate, set spindle speed, and tool number. Subsequently, the monitoring system calls the baseline correlation model according to the node type in the causal directed graph to generate predicted values corresponding to the current observations.
[0040] The predicted value represents the node value that should occur under normal operating conditions, current process conditions, and the current program segment. In this embodiment, the deviation vector refers to the set of deviation items arranged in node order, with each deviation item corresponding to the abnormal characterization result of a node at the current moment. To obtain the deviation items, the current observed value and the predicted value are compared item by item. The difference direction indicates whether the current observed value is higher or lower than the predicted value; the difference magnitude indicates the degree of deviation, which can be the absolute difference between the current observed value and the predicted value, or it can be normalized by combining the normal fluctuation range of the corresponding node; the difference duration indicates the length of time a certain deviation state is continuously maintained, which the monitoring system can accumulate by observing the deviation state of the same node through multiple consecutive time windows. If a node experiences a momentary deviation within a single time window, but returns to normal in the next time window, the deviation item of that node can be recorded only as a short-term disturbance and not involved in subsequent main inference. If a node maintains deviation in the same direction for multiple consecutive time windows, and the deviation magnitude exceeds the node's abnormal judgment threshold, then the deviation item of that node is marked as valid abnormal evidence. The anomaly detection threshold can be determined based on the upper limit of normal fluctuations for that node in historical normal processing data; different thresholds can be used for different nodes. After calculating the deviation terms for all nodes, the deviation terms are arranged according to the node order in the causal directed graph to form a deviation vector. This deviation vector retains the deviation direction, degree of deviation, and persistence characteristics, while maintaining a one-to-one correspondence with the nodes in the causal directed graph, facilitating its direct use in subsequent symptom node identification and anomaly root cause inference.
[0041] Probabilistic reasoning on a causal directed graph includes: identifying at least one of the processing progress node representing the deviation of the processing progress from the progress reference baseline and the alarm code node representing the alarm code as a symptom node; determining the anomalous probability of each node in the causal directed graph based on the deviation vector; and determining the root cause node and the anomalous propagation path based on the anomalous probability of each node and the probability contribution of each node to the symptom node.
[0042] In one embodiment, this embodiment further defines a probabilistic reasoning process on a causal directed graph, relative to the foregoing, for locating the root cause node of anomalies and determining the path of anomaly propagation based on the causal transmission relationship between nodes using the deviation vector.
[0043] Specifically, the monitoring system first determines symptom nodes based on the trigger source. When anomaly analysis is triggered by progress deviation, the processing progress node representing the abnormal duration of the current program segment or the cumulative completion progress lag is identified as the symptom node; when anomaly analysis is triggered by alarm codes, the alarm code node corresponding to the alarm event is identified as the symptom node; when both types of triggers occur simultaneously, multiple symptom nodes can be identified at the same time. After determining the symptom nodes, the monitoring system reads the deviation terms corresponding to each node in the deviation vector and performs a layer-by-layer backtracking analysis in conjunction with the edge directions in the causal directed graph. The anomaly probability is used to characterize the likelihood of a node being in an abnormal state. The calculation focuses on three aspects: first, the magnitude and duration of the node's deviation term—the larger the deviation and the longer it lasts, the higher the anomaly probability; second, the path length and edge connection relationship between the node and downstream symptom nodes—the shorter the path and the more direct the connection, the stronger the node's explanatory power for the symptom; and third, the chronological order of node deviations—upstream deviations that occur earlier are more likely to be the root cause than downstream deviations that occur later.
[0044] To avoid misidentifying symptom nodes as the root cause, the monitoring system prioritizes upstream process execution nodes and equipment status nodes during backtracking, checking whether the deviations of these nodes occurred earlier than the symptom node. If a path contains discontinuous edges, reversed time sequences, or intermediate nodes that remain normal for an extended period, that path is not considered a valid propagation path. If multiple upstream nodes have high anomaly probabilities, they are ranked according to their anomaly probability and their probability contribution to the symptom node. The probability contribution characterizes the explanatory strength of a node along a directed path in contributing to the current anomalous state of the symptom node, and can comprehensively consider the node's anomaly probability, path continuity, and the deviation strength of key nodes in the path. After ranking, the node with the highest contribution is selected as the root cause node, and the path from the root cause node to the symptom node with the highest contribution is selected as the anomaly propagation path. When multiple nodes have similar contributions, multiple candidate root cause nodes can be output simultaneously, along with their corresponding propagation paths, for subsequent verification by personnel.
[0045] Output the root cause node and the propagation path of the anomaly, including: determining the category of the root cause node based on the node type to which it belongs; generating handling suggestions based on the category of the root cause node; and recording the root cause node, the propagation path, and the handling suggestions.
[0046] In one embodiment, this embodiment further defines the output method of the abnormal root cause node and abnormal propagation path, relative to the foregoing, for converting the reasoning results into executable disposal information and forming a traceable event record.
[0047] Specifically, after obtaining the root cause node and propagation path of the anomaly, the monitoring system first generates handling suggestions based on the category to which the root cause node belongs. The category to which the root cause node belongs can be determined based on the node type and the corresponding variables. For example, process execution nodes may correspond to program setting mismatch, inconsistent tool parameters, or abnormal switching conditions; equipment status nodes may correspond to abnormal spindle load, servo response lag, abnormal vibration, or abnormal temperature rise. For different categories, the system pre-establishes a handling suggestion rule table, which records the mapping relationship between node categories and suggested actions. When the root cause node is a process execution node, the handling suggestions may include reviewing program segment parameters, reviewing tool call relationships, and verifying the process version; when the root cause node is an equipment status node, the handling suggestions may include checking spindle load changes, checking feed axis status, checking cooling and lubrication status, arranging spot checks, or pausing machining for review.
[0048] To facilitate operators' rapid understanding of the propagation process, the monitoring system can also output key nodes in the anomaly propagation path, the time of node occurrence, and the direction of node deviation in sequence. This allows operators to see how the anomaly gradually propagates from upstream causes to progress anomalies or alarm events. After generating the results, the system writes the root cause node, anomaly propagation path, handling suggestions, trigger type, trigger time, corresponding program segment number, equipment number, and processing task number into the event log. The event log is used for subsequent queries, statistics, and manual verification. If the operator adds a verification conclusion after handling the issue, such as confirming that the root cause is tool wear, spindle overload, or incorrect process parameter settings, the system can append this verification conclusion to the same event log as reference information for subsequent analyses of similar anomalies. In this way, the anomaly analysis results can not only be used for alarm interpretation and handling guidance at the current moment but also be reused as experience in subsequent similar tasks, thus forming a complete, verifiable, and traceable closed-loop processing chain for the anomaly monitoring process.
[0049] like Figure 2 As shown, a CNC equipment intelligent monitoring system based on machining progress is used to implement a CNC equipment intelligent monitoring method based on machining progress. The system includes: The data acquisition module is used to acquire progress data, equipment status data, process execution data, and alarm codes of the CNC equipment, and to determine the progress reference benchmark corresponding to the program segment based on historical normal machining data. The progress data includes the actual execution time and cumulative completion progress of the program segment, while the progress reference benchmark includes the reference execution time and reference completion progress of the program segment. The data acquisition module consists of an industrial communication interface unit, a signal acquisition unit, a CNC interface unit, and a storage unit. The industrial communication interface unit is used for communication with the CNC system, servo drivers, and spindle control devices; the signal acquisition unit is used to acquire equipment status signals; the CNC interface unit is used to read the program segment number, process execution information, and alarm codes; and the storage unit is used to store historical normal machining data and support the determination of the progress reference benchmark corresponding to the program segment.
[0050] The data fusion module is used to time-align progress data, equipment status data, process execution data, and alarm codes, and associate them according to program segment numbers to obtain a fused dataset. The data fusion module consists of an edge processor, a time synchronization unit, and a caching unit. The edge processor is used to time-align progress data, equipment status data, process execution data, and alarm codes, and associate them according to program segment numbers; the time synchronization unit is used to unify the time base of different data sources; and the caching unit is used to temporarily store multi-source data within the same sampling period to form a fused dataset.
[0051] The model building module is used to construct a causal directed graph, including process execution nodes, equipment status nodes, processing progress nodes, and alarm code nodes, based on the fused dataset. It then defines the edge directions in the causal directed graph based on processing sequence constraints and establishes a baseline correlation model. The model building module consists of an industrial computing unit, a model generation unit, and a model storage unit. The industrial computing unit performs graph structure analysis on the fused dataset; the model generation unit constructs the causal directed graph, defines the edge directions in the causal directed graph based on processing sequence constraints, and establishes a baseline correlation model based on historical normal processing data.
[0052] The inference output module is used to determine the deviation vector based on the current observed values and the predicted values corresponding to the benchmark correlation model when a deviation from the processing progress is detected or an alarm code is issued. Based on the deviation vector, it performs probabilistic inference on the causal directed graph to determine and output the root cause node and the anomaly propagation path. The inference output module consists of an inference processing unit, a display output unit, and a communication unit. The inference processing unit determines the deviation vector based on the current observed values and the predicted values corresponding to the benchmark correlation model when a deviation from the processing progress is detected or an alarm code is issued, and performs probabilistic inference based on the deviation vector. The display output unit outputs the root cause node and the anomaly propagation path. The communication unit sends the root cause node, the anomaly propagation path, and related handling information.
[0053] It should be noted that, in this document, the terms "comprising," "including," and any other variations are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Specific examples have been used in this document to illustrate the principles and implementation methods of the technical solutions of this application. The above examples are only for the purpose of helping to understand the methods and core ideas of this application. The above descriptions are merely preferred embodiments of this application. It should be pointed out that, due to the limitations of written expression and the objective existence of infinite specific structures, those skilled in the art can make several improvements, modifications, or changes without departing from the principles of this application, and can also combine the above technical features in an appropriate manner; these improvements, modifications, changes, or combinations, or the direct application of the concept and technical solutions of this application to other situations without modification, should all be considered within the scope of protection of this application.
Claims
1. A method for intelligent monitoring of CNC equipment based on machining progress, characterized in that, The method includes: Acquire the progress data, equipment status data, process execution data and alarm codes of the CNC equipment, and determine the progress reference benchmark corresponding to the program segment based on historical normal processing data. The progress data includes the actual execution time and cumulative completion progress of the program segment, and the progress reference benchmark includes the reference execution time and reference completion progress of the program segment. The progress data, equipment status data, process execution data, and alarm codes are time-aligned and associated according to program segment numbers to obtain a fused dataset. Based on the fused dataset, a causal directed graph including process execution nodes, equipment status nodes, processing progress nodes, and alarm code nodes is constructed, and the edge directions in the causal directed graph are limited based on the processing sequence constraints to establish a baseline association model. When the processing progress deviates from the progress reference benchmark or the alarm code is detected, the deviation vector is determined based on the current observation value and the predicted value corresponding to the benchmark association model. Based on the deviation vector, probabilistic reasoning is performed on the causal directed graph to determine and output the abnormal root cause node and abnormal propagation path.
2. The method according to claim 1, characterized in that, The historical normal processing data refers to historical samples that are consistent with the equipment model, processing technology and workpiece type of the current processing task and whose processing results are normal.
3. The method according to claim 2, characterized in that, The time alignment and association based on program segment number include: The progress data, equipment status data, process execution data, and alarm codes are timestamped and resampled according to a unified time reference. Establish the correspondence between the corrected data and the program segment based on the program segment number; The associated data is organized according to the equipment number, processing task number, program segment number, and time window to obtain the fused dataset.
4. The method according to claim 1, characterized in that, The processing sequence constraint is used to limit the edge direction of the causal directed graph, so that the process execution node points to the equipment status node, and the equipment status node points to the processing progress node or alarm code node.
5. The method according to claim 4, characterized in that, The establishment of the benchmark correlation model includes: The upstream node of each node is determined based on the causal directed graph. Using the data corresponding to the upstream node of each node as input and the value of each node in the historical normal processing data as output, the prediction relationship of each node is established respectively. The baseline correlation model is established based on the predicted relationships of each node.
6. The method according to claim 1, characterized in that, A deviation from the progress reference baseline is detected, including at least one of the following: The actual execution time of the detected program segment is greater than the reference execution time of the program segment; The cumulative progress of testing is lower than the reference completion progress.
7. The method according to claim 5, characterized in that, The determination of the deviation vector includes: Extract the current observation value from the fused dataset corresponding to the current moment; The predicted value corresponding to the current observation value is determined based on the benchmark correlation model; The deviation vector is generated based on the direction, magnitude, and duration of the difference between the current observation and the predicted value.
8. The method according to claim 7, characterized in that, The probabilistic reasoning on the causal directed graph includes: At least one of the processing progress node representing the deviation of the processing progress from the progress reference benchmark and the alarm code node representing the alarm code is identified as a symptom node; The anomaly probability of each node in the causal directed graph is determined based on the deviation vector. The root cause node and the propagation path of the anomaly are determined based on the anomaly probability of each node and the probability contribution of each node to the symptom node.
9. The method according to claim 8, characterized in that, The output of the root cause node of the anomaly and the anomaly propagation path includes: The category of the abnormal root cause node is determined based on the node type to which the abnormal root cause node belongs; Based on the category of the abnormal root cause node, a handling suggestion is generated; Record the root cause node of the anomaly, the propagation path of the anomaly, and the handling suggestions.
10. A CNC equipment intelligent monitoring system based on machining progress, used to implement the CNC equipment intelligent monitoring method based on machining progress as described in any one of claims 1-9, characterized in that, The system includes: The data acquisition module is used to acquire the progress data, equipment status data, process execution data and alarm codes of the CNC equipment, and determine the progress reference benchmark corresponding to the program segment based on historical normal machining data. The progress data includes the actual execution time and cumulative completion progress of the program segment, and the progress reference benchmark includes the reference execution time and reference completion progress of the program segment. The data fusion module is used to perform time alignment on the progress data, the equipment status data, the process execution data, and the alarm codes, and to associate them according to the program segment number to obtain a fused dataset; The model building module is used to construct a causal directed graph including process execution nodes, equipment status nodes, processing progress nodes and alarm code nodes based on the fused dataset, and to limit the edge directions in the causal directed graph based on the processing sequence constraints to establish a baseline association model. The inference output module is used to determine the deviation vector based on the current observation value and the predicted value corresponding to the benchmark association model when the processing progress deviates from the progress reference benchmark or the alarm code is detected, and to perform probabilistic inference on the causal directed graph based on the deviation vector to determine and output the abnormal root cause node and abnormal propagation path.