Processing quality virtual measurement and simulation method based on ubiquitous perception

Abstract

The invention relates to the field of computer data processing and information retrieval, and particularly discloses a machining quality virtual measurement and simulation method based on ubiquitous perception. The method comprises the steps that part state data, equipment state data and sensor timestamps of a machining production line are acquired, unified time reference configuration, drift correction and channel denoising processing are carried out, and an alignment segment data structure is generated; extracting channel, geometric measuring point, dynamics and motion control features, and establishing a process feature element set structure; a virtual measurement model is constructed based on the structure, simulation measurement, space registration and layered rendering are executed according to the target surface area, the sampling path and the attitude parameters, and three-dimensional chromatographic display data are generated; and finally, outputting a parameter correction instruction to realize dynamic updating of the model. According to the method, ubiquitous perception and virtual measurement are fused, so that visual, precise and intelligent evaluation of the machining process is realized, and the quality control efficiency and prediction precision under complex working conditions are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of computer data processing and information retrieval, and in particular to a method for virtual measurement and simulation of processing quality based on ubiquitous perception. Background Technology

[0002] In the field of computer data processing and information retrieval, existing solutions for processing part status data, equipment status data, and sensor timestamps from production lines typically employ a distributed acquisition, offline alignment, and independent modeling process per device. This approach suffers from limitations such as incomplete unified time reference configuration and drift correction, non-standardized handling of channel denoising and step size resampling, marking missing intervals without valuation, and discontinuous segmentation of cross-device clock alignment, event window truncation, and operational mode labeling. Existing methods often rely on fixed process route list mappings and manually set alignment strategies. In scenarios involving segmented data structures organized by fragments, these methods are prone to inconsistencies in channel standardization, geometric measurement point, dynamics, and motion control feature extraction and cross-domain alignment fusion, as well as incomplete element numbering and element-data mapping. Consequently, they struggle to achieve stable virtual measurement model assembly and simulation measurements based on target surface areas, sampling paths, and attitude parameters. For the joint processing of 3D chromatographic display data with geometric deviation comparison, region clustering and error assessment, equipment health status mapping and trend analysis, and working condition matching to generate parameter correction instructions, existing technologies generally have shortcomings such as insufficient synchronization, unclear judgment boundaries and inconsistent parameter transmission in the association between spatial registration, color mark encoding and layered rendering and model update instruction structure. It is difficult to form a continuous process of acquisition and alignment, feature extraction, virtual measurement and display, comparison and assessment and generation of model update instruction structure in the processing production line scenario, resulting in unstable association between model update instruction structure and the aforementioned data and rules and poor process connection. Summary of the Invention

[0003] To address the aforementioned technical problems, this invention provides a method for virtual measurement and simulation of processing quality based on ubiquitous sensing, comprising: Acquire part status data, equipment status data and sensor timestamps from the processing production line, perform unified time base configuration, drift correction, channel denoising and step size resampling, mark missing intervals without estimating, cross-device clock alignment and event window truncation and operation mode label segmentation, and generate an aligned segmented data structure organized by segments; The aligned segmented data structure is obtained, and channel standardization, geometric measurement points, dynamics, motion control feature extraction and cross-domain alignment fusion, feature numbering and feature-data mapping processing are performed to generate a process feature feature set structure. Based on the structure of the process feature set, virtual measurement model assembly, simulation measurement based on target surface area, sampling path and attitude parameters, spatial registration, color mark encoding and layered rendering are performed to generate three-dimensional color display data. Based on the three-dimensional chromatographic display data, geometric deviation comparison and region clustering, error evaluation and equipment health status mapping and trend analysis and working condition matching are performed to generate parameter correction instructions and obtain the model update instruction structure.

[0004] Furthermore, the part status data, equipment status data, and sensor timestamps from the processing production line include: The part status data of the processing line includes the process progress of the part at the workstation, the clamping posture and number association, as well as design tolerance references and measurement point location references; Equipment status data includes machine tool spindle speed, feed rate, tool number, servo load, temperature and vibration monitoring readings, and records its data identifier, unit, sampling period, trigger conditions and abnormal code in the channel description to support cross-equipment standardization and operation mode label segmentation; Sensor timestamps are generated by edge acquisition nodes at the moment of each sampling and appended to the original measurement value, subject to a unified time reference configuration.

[0005] Furthermore, the process of generating aligned segmented data structures organized by fragments also includes: Acquire part status data, equipment status data and sensor timestamps from the processing production line, establish channel descriptions through the fieldbus gateway and data access module, load NTP and PTP dual-stack alignment and drift correction with unified time reference configuration, and only mark time anomalies without rewriting them to generate a perception dataset organized by workstation, equipment and part number. The channel list, sampling step size and missing segment label are extracted from the perception dataset. Denoising is performed by selecting a rule base strategy based on channel type, resampling at the target time axis step size and only labeling the missing interval without estimating the value, and a structured time series with a three-level index and a missing segment dictionary is generated. The alignment mapping table anchors are obtained from the structured time series. Cross-device clock alignment, event window truncation based on joint thresholds and anchors, and tag segmentation processing according to the operation mode dictionary are performed to generate an aligned segmented data structure containing an event window list and an operation mode fragment list.

[0006] Furthermore, the process of generating the structure of the process feature set also includes: Based on the aligned segmented data structure, channel standardization is performed with unified units, unidirectional direction and consistent boundaries. Then, equipment instance mapping is performed according to the rules of equipment number, channel source and workstation binding to obtain standardized instance data organized by operation mode segments. Geometric measurement point features, dynamic features, and motion control features are extracted from standardized instance data. Temporal alignment based on alignment anchor points and spatial alignment based on target surface and measurement point references are performed. Cross-domain fusion processing is carried out by segment aggregation to generate a fusion feature set containing dual indices of measurement points and path segments. The feature set is used to obtain the feature type, boundary, association and supporting evidence. The process features are then processed by bucketing by process and constructing candidate feature chains based on path continuity, as well as by hierarchical numbering and feature-data mapping relationship construction, to generate the process feature feature set structure.

[0007] Furthermore, the process of generating three-dimensional chromatographic display data also includes: Based on the structure of the process feature element set, the model component library is retrieved, the local workspace is constructed, the sampling strategy is implemented, the rule binding of posture and path constraints is performed, and the assembly drawing topology is processed to obtain the virtual measurement model. The target surface area, sampling path and attitude parameters are extracted from the virtual measurement model, and the simulation measurement operation and element and process level result aggregation processing are performed according to the assembly drawing execution sequence to generate virtual measurement results. The registration geometry and fragment merging records are obtained from the virtual measurement results. The spatial registration is driven by the reference datum, the color code is selected according to the tolerance limit, and the layered rendering process based on the feature color layer and annotation layer is performed to generate three-dimensional color display data.

[0008] Furthermore, the process of obtaining the model update instruction structure also includes: Based on 3D chromatographic display data, geometric deviation comparison under design tolerance datum and assembly drawing constraints is performed, common area coverage records are disambiguated, and spatial clustering by curvature and neighborhood is performed. Combined with semantic aggregation processing of element number and path segment, an error label sequence is obtained. Feature segments, equipment operation segments, and tool usage segments that match the label boundaries are extracted from the error label sequence. Error judgment and equipment health status mapping are performed based on the basic and combined entries of the rule base to generate error judgment results. The error evaluation results are used to obtain label-attribution pairs and fragment cross-references. Trend analysis of batch timelines and spatial grids is performed, and working conditions are matched according to operating mode, equipment status, and tool stage. The process of re-injectable time base fine-tuning, sampling path refinement, and attitude constraint adjustment update items is output to generate model update instruction structure.

[0009] Furthermore, the process of running the virtual measurement model also includes: After reading the topological order in the assembly drawing, an execution sequence is generated according to the combination strategy of parallel measurement and serial measurement. The target surface region is extracted from the local workspace description. The target surface region is used to limit the spatial range of the simulation measurement, including the target surface boundary, the adjacent transition area and the safety clearance area.

[0010] Furthermore, on the same element component, sampling paths and attitude parameters are extracted by binding records according to rules. The sampling path is a discrete sequence of feasible paths within the target surface area, and the attitude parameters are the detection attitude or sensing attitude when executing along the path. Together, they constitute the motion and observation configuration of the simulation measurement.

[0011] Furthermore, during the operation preparation phase, path constraints are aligned with equipment travel, acceleration / deceleration capabilities, and tool change timing. When travel or synchronization requirements are not met, the path is segmented and rearranged, and the rearrangement record is written into the execution sequence notes.

[0012] Furthermore, in the result aggregation stage, sampling points and corresponding attitudes are collected for each element component to form component-level measurement segments. At the element level, the segments are spatially reordered and merged. The reordering follows the natural traversal order of the target surface boundary and the stability priority principle of attitude switching. The segment merging is carried out under the condition that the boundary is continuous and the criteria are consistent.

[0013] The key innovations of this invention include: (1) In the process of acquiring part status data, equipment status data and sensor timestamps of the processing production line, a continuous link is proposed, which includes unified time reference configuration, drift correction, channel denoising and step size resampling, marking missing intervals without valuation, cross-device clock alignment and event window truncation and segmented processing of operation mode labels, and generates an aligned segmented data structure organized by segments.

[0014] (2) After obtaining the aligned segmented data structure, construct a structured organization method for channel standardization, geometric measurement points, dynamics, motion control feature extraction and cross-domain alignment fusion, element numbering and element-data mapping processing, output the process feature element set structure and solidify the element-data mapping relationship.

[0015] (3) Based on the process feature element set structure, a model-display integrated process is formed, which includes virtual measurement model assembly, simulation measurement based on target surface area, sampling path and attitude parameters, spatial registration, color mark encoding and layered rendering processing. It is then connected with geometric deviation comparison and region clustering, error evaluation and equipment health status mapping and trend analysis and working condition matching to generate parameter correction instruction processing closed loop to obtain the model update instruction structure.

[0016] The following are its main beneficial effects: (1) It acts on the formation link of the aligned segmented data structure. By marking missing intervals without valuation and coordinating cross-device clock alignment, it establishes verifiable time and semantic boundaries, reduces the dependence on interpolation and manual alignment strategies, and is suitable for multi-source data processing scenarios organized by segments.

[0017] (2) It acts on the construction process of the process feature element set structure. Through channel standardization and geometric measurement points, dynamics, motion control features extraction and cross-domain alignment fusion, combined with element numbering and element-data mapping processing, it realizes traceable organization from data to elements, reduces the dependence on manual mapping and one-time rules in the existing scheme, and is suitable for multi-process and multi-equipment collaboration scenarios.

[0018] (3) It operates on the virtual measurement to display and evaluation to instruction generation chain. Through virtual measurement model assembly and spatial registration, color mark encoding and layered rendering, it links geometric deviation comparison and region clustering, error evaluation and equipment health status mapping and trend analysis and working condition matching to generate parameter correction instruction processing, forming a continuous process from simulation measurement to model update instruction structure. It is suitable for online decision-making and batch iteration scenarios in the processing production line. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating a virtual measurement and simulation method for processing quality based on ubiquitous sensing, provided in an embodiment of this application. Detailed Implementation

[0020] Example 1: Refer to Figure 1 This is a flowchart illustrating a virtual measurement and simulation method for processing quality based on ubiquitous sensing, provided by an embodiment of the present invention. The process may include at least steps S100-S400: S100: Acquire part status data, equipment status data and sensor timestamps from the processing production line, perform unified time reference configuration, drift correction, channel denoising and step size resampling, mark missing intervals without estimating, cross-device clock alignment and event window truncation and operation mode label segmentation, and generate an aligned segmented data structure organized by segments. S200: Obtain the aligned segmented data structure, perform channel standardization, geometric measurement point, dynamics, motion control feature extraction and cross-domain alignment fusion, element numbering and element-data mapping processing, and generate the process feature element set structure; S300, based on the process feature element set structure, performs virtual measurement model assembly, simulation measurement based on target surface area, sampling path and attitude parameters, spatial registration, color mark encoding and layered rendering processing to generate three-dimensional color display data; S400, based on three-dimensional chromatographic display data, performs geometric deviation comparison and region clustering, error assessment and equipment health status mapping and trend analysis and working condition matching, and generates parameter correction instruction processing to obtain the model update instruction structure.

[0021] Step S100 includes at least steps S110-S130: S110. Acquire part status data, equipment status data and sensor timestamps from the processing production line, perform unified time reference configuration and drift correction processing, and obtain a sensing dataset. In this invention, the part status data, equipment status data, and sensor timestamps of the processing production line specifically refer to: Part status data, which is semantic information oriented towards processes, accessed by the field measurement and control network via a fieldbus gateway and data access module, and organized by workstation, equipment, and part number. It includes the part's process progress at the workstation, clamping posture, and number association, as well as reference fields such as design tolerance references and measurement point location references for subsequent element mapping and virtual measurement; Equipment status data, which is the machining process operation quantity associated with the above-mentioned part status data on the same time axis, including at least machine tool spindle speed, feed rate, tool number, servo load, temperature, and vibration monitoring readings, and records its data identifier, unit, sampling period, trigger condition, and abnormal code in the channel description to support cross-equipment standardization and operation mode label segmentation; Sensor timestamps are generated by the edge acquisition nodes at the moment of each sampling and appended to the original measurement value, subject to a unified time base. Under the constraints of quasi-configuration, a high-precision master clock is prioritized, and NTP / PTP dual-stack alignment is provided. For older models of equipment without external time synchronization capabilities, round-trip delay is estimated through software heartbeat ranging and converted into local correction deviation, which is then written into the channel description. At the same time, a time deviation queue is maintained during operation to trigger drift detection and clock fine-tuning. When timestamp rollback, duplication, or significant jumps are encountered, the original data is not rewritten but marked as time anomalies, and the start and end positions of the anomaly segment are bound. After the above access, reference configuration, and drift correction, the part status data, equipment status data, and aligned sensor timestamps are associated on the same time axis to form a unified input object that is serializable, replayable, and traceable. Delayed messages are marked with placeholders according to the delay tolerance time in the channel description. The aggregated product is the sensing dataset, which serves as the time alignment prerequisite for subsequent channel denoising, step size resampling, missing segment marking, and the establishment of standardized instance data by the S200 module.

[0022] Specifically, the existing measurement and control network deployed in the production line is used as the input channel. Part status data, equipment status data, and sensor timestamps are accessed via a fieldbus gateway and data access module. The part status data indicates the part's process progress, clamping posture, number association, design tolerance reference, and measurement point location reference at the workstation. The equipment status data includes machine tool spindle speed, feed rate, tool number, servo load, and temperature and vibration monitoring readings. The sensor timestamps are generated by edge acquisition nodes at the moment of sampling and appended to the original measurement value. During the input access phase, the data access module identifies and parses message formats from different sources, establishing channel descriptions for three types of carriers: message queues, file streams, and field storage areas. The channel descriptions record the corresponding data identifier, unit, sampling period, trigger conditions, and exception code, enabling channel-level operations in subsequent processing links. Furthermore, the time reference service loads a unified time reference configuration during system initialization. This unified time reference configuration specifies the time source priority, alignment threshold, jump suppression time window, and daemon thread cycle. The time source priority is set in the order of high-precision master clock first, followed by local reference. To achieve cross-device alignment, the time reference service provides dual-stack support for both Network Time Protocol (NTP) and Precision Time Protocol (PTP), selecting the appropriate protocol channel based on device capabilities. For older acquisition devices lacking external clock synchronization capabilities, the time reference service estimates round-trip delay using software heartbeat ranging and converts the estimate into a local correction deviation, which is then written into the corresponding channel description. For drift correction, the system continuously maintains a time deviation queue for each channel during operation. When the queue length meets the minimum sample requirement in the unified time reference configuration, a drift determination process is triggered. This process iterates through the channel descriptions, constructs a drift trend based on the time deviation queue, and if the trend exceeds the alignment threshold, the clock fine-tuning interface is called to incrementally correct subsequent timestamps for that channel, recording the difference before and after correction, along with the trigger time, as audit records for subsequent traceability. In terms of anomaly handling, when timestamp rollback, duplication, or significant jumps are detected, the time reference service marks the sampling point as a time anomaly without modifying the original measurement value, and binds the start and end positions of the anomaly segment to the channel description. Data within the anomaly segment is not deleted and continues to enter the subsequent link along with the channel data. After completing the above unified time reference configuration and drift correction processing, the data access module repackages messages from different devices, associating part status data, equipment status data, and aligned sensor timestamps on the same time axis to form a unified input object for subsequent processing. The unified input object is organized by workstation, equipment, and part number.Understandably, within the same batch after access is completed, sampling gaps and delayed arrival messages are still allowed. The data access module uses the delay tolerance time in the channel description to mark the unarrived messages as placeholders, and the marking information will be identified and processed in subsequent sections. The output of the above process is denoted as the sensing dataset. The sensing dataset contains the channel-level sample value sequence, the reference time after unified time base configuration, the correction record generated by drift correction processing, and the abnormal segment marking. This sensing dataset is passed to the next section of this main step for use. In the subsequent S120, it is directly consumed as the sensing dataset. At the same time, in the entire process, this sensing dataset is also the time alignment prerequisite for the subsequent S200 module to establish standardized instance data.

[0023] During operation, the unified time reference configuration is maintained by the system configuration management module. This module sets a hot-update strategy for configuration changes, stipulating that switching only occurs at sampling boundaries to avoid time discontinuities caused by reference switching within the same sampling frame. For multi-segment distributed production lines, the time reference service deploys slave clock instances in each segment. These slave clock instances are synchronized with the master clock instance and take over the reference provision responsibility in case of a segment failure. Once the network recovers, the master clock instance reverts to a high-priority source, and drift correction is reset to normal window parameters to prevent the cumulative deviation impact of fault recovery. For newly accessed devices, the system automatically assigns channel description entries during the registration phase and reads the device's self-reported capability set. If the capability set includes hardware timestamp support and an external trigger interface, the time reference service enables hardware trigger mode, aligning the sensor timestamp with the production line beat generator at the hardware level. If the device only supports software sampling, the time reference service marks it as a software-aligned channel and provides approximate alignment using heartbeat ranging and delay estimation. After the edge node fault is recovered, the time reference service generates a connection record by comparing the last timestamps of the channels before and after recovery. This connection record will be identified and retained in the missing segment marking process in subsequent sections. In summary, the perception dataset formed through access, reference configuration, and drift correction already possesses the functional characteristics of being serializable, replayable, and traceable in its data structure. Subsequent sections will build upon this foundation to carry out specific channel denoising, step size resampling, and missing segment marking.

[0024] S120. Extract the channel list, sampling step size, and missing segment labels from the perception dataset, perform channel denoising and step size resampling, and label the missing intervals without estimating the value, to generate a structured time series. Specifically, the aforementioned sensing dataset is used as the sole input source. This sensing dataset is output by S110 and already contains channel descriptions, aligned sensor timestamps, and anomaly segment annotations. At the beginning of processing, the system extracts a channel list from the sensing dataset. This list records the identifier, source device, measured physical quantity, unit, and original sampling step size for each physical or logical channel. Simultaneously, the system parses the time scale in the sensing dataset and reads the actual sampling step size distribution for each channel to determine if sampling jitter or gaps exist. For the channel denoising process, the system selects rules from the denoising strategy library based on the type of measured physical quantity in the channel list. The rule content is maintained by a knowledge base, which binds the applicable channel type, frequency band range, and anomaly flag interaction conditions to each rule during import. The denoising process employs a pipelined approach. First, it performs spike suppression and glitch identification. Spike suppression is triggered by abnormal code points or time anomaly markers in the channel description. When the difference between a sampling point and its neighbors exceeds the spike amplitude threshold set by the rule base, that point is marked as a spike sample, but the sample value is not modified. Next, it performs smoothing filtering. The filtering window length is selected based on the channel type, adhering to the constraint of not excessively altering the actual trend of change. Specifically, a short window is used for cutting vibration channels, and a long window is used for slow temperature variable channels. The processed samples are logged with the original value and the difference before and after processing. Furthermore, to avoid phase lag during denoising, the system uses a symmetrical sliding window approach. If the window cannot completely cover the sequence endpoints, it reverts to an endpoint safety policy, which only records the denoising intent without modifying the endpoint samples. For channels with dropped frames or delayed arrival, the denoising process does not attempt interpolation, preserving the original gap attributes to provide a basis for subsequent step-size resampling and missing interval labeling.

[0025] During step-size resampling, the system reads the original sampling step size for each channel from the channel list and combines it with the target time step size in the unified time reference configuration to construct a target time axis. The target time axis covers the start and end times within the processing window. For different channel alignment requirements, the system provides two resampling methods: a strict alignment method for channels that need to be consistent with the device control cycle, and a loose alignment method for monitoring channels that only need to be roughly aligned with the unified time reference. The strict alignment method requires alignment to the original sample or the use of placeholders to represent gaps at each time scale of the target time axis. The loose alignment method searches for the nearest original sample within the allowed drift range as a representative value, and records the difference between the original time and the target time of the representative value in the resampling supplementary information. It is important to emphasize that this section follows the principle of only marking missing intervals without assigning estimates. For any data point on any time scale where no original sample exists, the system does not perform interpolation calculations or assign an estimated value, but instead outputs a missing segment marker and a placeholder symbol. The placeholder symbol is distinguished from the actual sample value in the data structure and is identified as an object that cannot participate in aggregation in subsequent calculations. The sources of missing segment markers include abnormal segment markers in S110, device connection records, and network latency trigger records. The system summarizes these sources into a missing segment dictionary. The missing segment dictionary clarifies the start and end times and cause categories of the missing interval for each channel on the target time axis. The cause categories are used for tracking in subsequent fault attribution and quality assessment.

[0026] After channel denoising and step-size resampling are completed, the system enters the structured organization stage. This stage packages the resampled sequences of each channel and their corresponding metadata into a unified structured time series. This structured time series is expressed as a set of frames indexed by a time scale. Each frame contains sampled values ​​from multiple channels, the corresponding alignment time, missing segment markers, and channel metadata references. To accommodate subsequent cross-device and cross-workstation processing, the structured organization stage establishes a three-level index based on workstation, device, and part number. This allows channel data for the same part at different workstations to be concatenated, and channel data for the same device in different batches to be aggregated. Regarding anomaly handling, if a channel is found to be completely missing within a full window during resampling, the system marks that channel as completely missing and retains its metadata and null placeholders in the structured time series for subsequent determination of whether the influence of that channel needs to be removed in the S130 operation mode label segmentation processing. The output after the above processing is a structured time series. The structured time series is directly used in the next subsection of this main step and is consumed as the input of the structured time series in S130. At the same time, the structured time series will be referenced by the channel normalization and device instance mapping processing of S210 in the data stream across the main steps, as the original sequence basis for establishing the normalized instance data.

[0027] From an operation and maintenance perspective, the parameters and strategies for channel denoising, step-size resampling, and missing segment labeling are uniformly managed by a rule base. The rule base supports strategy switching based on production shift, workpiece material, and tool type. The switching timing is triggered by the scheduling module within the safety window of the process switching, preventing strategy mixing within the same processing segment. When a new channel is added to the sensing dataset, the system automatically retrieves the rule mapping for that channel in the next processing window. If the rule base does not contain the corresponding entry, the system sets the channel to the default strategy. The default strategy does not perform denoising or resampling, only aligns the time axis, and uses placeholders to represent unknown processing, thereby maintaining the stability and consumability of the structured time series format. For channels with frequent anomalies, the system records the time and type distribution of anomalies and enables observation mode in the rule base. In observation mode, only annotation and recording are performed on the channel; no modification operations are performed, facilitating the restoration of the channel's behavior trajectory in subsequent analysis.

[0028] S130. Perform cross-device clock alignment, event window truncation, and runtime label segmentation on the structured time series to generate an aligned segmented data structure; Specifically, the structured time series output by S120 is used as input. This structured time series has been resampled under a unified time reference and carries missing segment markers and a three-level index. During cross-device clock alignment, the system first locates the relevant channels of the same part at different workstations based on the three-level index in the structured time series, and reads the alignment time and resampling information attached to each channel during generation. The goal of cross-device clock alignment is to establish the time correspondence between channels without changing the original sampled values, so that the readings of different devices can be analyzed synchronously at the same time scale. To this end, the system constructs an alignment reference between devices. This alignment reference obtains the main reference time from the unified time reference configuration, and introduces the cycle boundary of the device control signal and the machine tool cycle start / end flag as auxiliary anchor points. The alignment process scans the set of structured time series frames sequentially within each processing window. When a beat boundary or cycle marker is encountered, a set of alignment anchor points is established, and the anchor point time is written to the alignment mapping table. For channels with slight drift, the system only records the anchor point offset in the alignment mapping table without modifying the original time, thus maintaining the consistency of marking missing intervals in S120 without estimating the value. In abnormal situations, such as when a device does not generate any beat or cycle marker within a window, the system degenerates to using a reference time with a unified time base as the anchor point and marks the no-anchor-point state in the alignment mapping table, prompting subsequent processing links to note that the time correspondence of this device has low reliability.

[0029] After completing cross-device clock alignment, the system enters the event window capture phase. Events captured in the event window are defined as identifiable stages or instantaneous behaviors related to the machining process, such as workpiece clamping completion, tool contact, sudden spindle loading changes, tool path transitions, and unloading. These events are not a fixed list but are determined jointly by multi-channel joint features in the structured time series and anchor points in the alignment mapping table. The system configures trigger rules for each event, consisting of channel thresholds, temporal relative relationships, and anchor point neighborhoods. During runtime, when a channel reading meets the threshold condition and the relative anchor point is within the specified neighborhood, the system determines that the event is valid, records the start and end frame indices of the event, and marks the set of frames within the corresponding time range as the event window. To handle the impact of missing segments on event determination, if the system encounters a placeholder within the event window, it retains and labels the frame using a weak sampling strategy. Weakly sampled frames are not used for computational aggregation but are retained in the window segment for reference during subsequent label segmentation and model calls. Once the event window is captured, each window is bound to a specific part, workstation, and equipment index, forming a set of processing segments that can be directly located and replayed in subsequent stages.

[0030] In the segmentation processing of operation mode labels, the system assigns labels and segments each event window based on a predefined operation mode dictionary in the equipment ledger and process route list. The operation mode dictionary consists of the names of the operation modes that the machining unit can execute, trigger conditions, termination conditions, and mutual exclusion relationships. Common operation modes include no-load, acceleration, constant speed cutting, variable speed cutting, tool change, shutdown, and standby. Within each event window, the system scans sequentially from the start frame to the end frame. When the trigger condition for a certain operation mode is met, the segment for that operation mode is started, until its termination condition is met or it is preempted by a mutually exclusive operation mode. For frames with missing segment markers, the system includes them in the current operation mode segment but adds a missing mask to indicate a conservative strategy for the numerical interpretation of that segment during subsequent evaluation. To avoid segment fragmentation caused by frequent switching of operation modes at boundaries, the system introduces a minimum segment length constraint and a jitter suppression threshold. A new segment is generated only when a candidate operation mode continuously meets the conditions and exceeds the minimum segment length; if the segment length is insufficient, the system merges the candidate segment into an adjacent stable segment and records the jitter merging. Ultimately, the operation mode label segmentation breaks down each event window into several segments with clear operation mode labels. Each segment is associated with part, workstation, equipment, alignment mapping, start and end frames, missing mask, and operation mode name.

[0031] After the three sub-processes described above are completed in series, the system integrates the results of cross-device clock alignment, event window capture, and operation mode label segmentation to generate an aligned segmented data structure. This aligned segmented data structure serves as a standard input format for subsequent feature extraction and process element extraction. Its core components include an alignment mapping table, an event window list, and an operation mode fragment list. These three components are linked through indexes and time ranges, enabling subsequent cross-domain feature alignment and fusion. The aligned segmented data structure is written to the data directory and registered with the process scheduler as a consumable field name during output, clarifying its destination in subsequent steps: This aligned segmented data structure is directly consumed by the channel standardization and device instance mapping processing of S210 as an aligned segmented data structure, and the scheduler uses the aligned segmented data structure as one of the input parameters when entering S210; at the same time, this data structure provides fragment boundaries and operating mode context for cross-domain feature alignment and fusion in S220 in more distant cross-master step scenarios, provides an event window organization method for process feature element extraction in S230, and when the process continues to advance to S300 and S400, the fragment index in the operating mode fragment list can also be used as a reference for simulation measurement sampling paths and boundary conditions for error attribution.

[0032] In summary, the technical effects of this step are as follows: Through unified time base configuration and drift correction processing in S110, channel denoising and step size resampling and missing interval marking without valuation in S120, and cross-device clock alignment, event window truncation and operation mode label segmentation processing in S130, a stable and traceable aligned segmented data structure is formed, providing consistent time and semantic boundaries for subsequent standardized instance data construction and feature fusion.

[0033] Step S200 includes at least steps S210-S230: S210. Obtain the aligned segmented data structure, equipment ledger and process route list, perform channel standardization and equipment instance mapping processing to obtain standardized instance data; Specifically, the aligned segmented data structure output from the previous step is used as the sole input for time and semantic boundaries. This aligned segmented data structure includes an alignment mapping table, an event window list, and a list of running mode fragments, and is associated with a three-level index of part number, workstation number, and equipment number. Simultaneously, the equipment ledger and process route list are loaded. The equipment ledger is a collection of master data for equipment on the production line, recording fields such as equipment model, serial number, sensor interface, physical channel range, control system version, and maintenance status. The process route list is part-oriented process arrangement data, recording process sequence, workstation allocation, tool list, target surface, measurement point references, and process parameter boundaries. During the input inspection phase, the system first performs an integrity check on the event window list and the running mode segment list of the aligned segmented data structure. The check logic is to compare each segment to see if its start and end frames are within the valid range of the alignment mapping table. When the start and end frames are found to be out of bounds or overlapping, the problematic segment is marked as a segment to be reviewed, retained but not involved in subsequent standardization calculations. For missing equipment numbers or deactivated markings in the equipment ledger, the system marks the corresponding channel as a frozen channel in the current batch, does not participate in instance mapping, and only outputs the frozen record in the results, which is convenient for subsequent handling in the process route revision stage. After input checks are completed, the system enters channel standardization processing. Channel standardization refers to unifying units, aligning directions, and ensuring dimensional annotation and boundary consistency for readings from different devices and channels without altering the original value source. Unit unification is performed according to the sensor interface definitions in the equipment ledger, rewriting the units of the same physical quantity on different devices back to the standard units, and recording the conversion relationship in the channel metadata. Alignment of directions is used for channels with symmetrical installations or opposite direction definitions, correcting for positive and negative directions using the installation field in the equipment ledger. Dimensional annotation writes the physical meaning, statistical attributes, and sampling form of each channel into the channel description notes, enabling subsequent feature calculations to select the appropriate processing path based on the annotations. Boundary consistency is based on the process parameter boundaries in the process route list, marking sampling points outside the specified range. Boundary marking does not modify the numerical value, but only adds a status to the sampling point for reference in subsequent decisions. For frames marked as weakly sampled or containing missing masks in the aligned segmented data structure, channel standardization only performs annotation and marking, without performing unit conversion or direction correction, to prevent the introduction of falsified processing information in the missing interval.

[0034] In the device instance mapping process, the system uses the operation mode segments with aligned segmented data structures as the outer loop and the device ledger as the device dictionary to map the multi-channel data within each operation mode segment to specific device instances. The mapping rules are constrained by three conditions: device number, channel source, and workstation binding. When the workstation where the operation mode segment is located matches the workstation binding in the device ledger, and the channel source equals the sensor interface registered in the device ledger, the channel is determined to belong to that device instance. If the same channel simultaneously meets the conditions of multiple devices, a unique device instance is selected according to the priority field in the device ledger, and the remaining candidate instances are recorded as redundant mappings for reference, for subsequent traceability. To address multi-device collaboration scenarios, the system allows a certain operation mode segment to be mapped to multiple device instances simultaneously, provided that their respective channel sets do not overlap or the overlap rate does not exceed a threshold. When an overlap exceeding the threshold occurs, the operation mode segment is split into multiple sub-segments. Each sub-segment retains only the channel set matching its device instance, and the source segment index is written into the segment metadata to achieve segment-level traceability. After instance mapping is completed, the system generates an instance context for each device instance within the current operating mode segment. The instance context includes the device model, control system version, channel list, effective frame range, and alignment anchor set. This context is referenced as an environment parameter in subsequent feature extraction and fusion. Regarding exception handling, if the device ledger lacks a channel definition for a certain model but a channel from that model appears in the alignment segment data structure, the system adds that channel to the pending filing list and records it with a temporary channel name, without interrupting the process. At this point, the output of this section is standardized instance data. This standardized instance data is organized according to operating mode segments and device instances, including channel sequences with unified units and directional alignment, boundary markers, instance context, and a list to be reviewed. This standardized instance data is directly consumed in the next section of this main step. In subsequent steps, geometric measurement point features, dynamic features, and motion control features are extracted from the standardized instance data, and cross-domain feature alignment and fusion are performed to generate a fused feature set used as input fields. Simultaneously, in the cross-main step data link, the standardized instance data also provides queryable device and segment context for the selection of target surfaces and sampling paths in the virtual measurement stage.

[0035] S220. Extract geometric measurement point features, dynamic features, and motion control features from standardized instance data, perform cross-domain feature alignment and fusion, and generate a fused feature set. Specifically, the aforementioned standardized instance data is used as the input source. This standardized instance data is generated in the previous section and includes multi-channel sequences and instance contexts organized by operating mode segments and equipment instances. The primary object of feature extraction is geometric measurement point features, which are defined as geometrically related quantities surrounding the target surface of the part and the measurement point position. These include measurement point positioning references, the spatial relationship between the measurement point and the tool position, the mapping of the measurement point in the workstation coordinates, and the correspondence between the measurement point and the process route list entries. The system first queries the measurement point references and target surface fields in the process route list and associates them with the instance context in the standardized instance data to obtain the set of measurement points potentially involved in each operating mode segment. After the association is completed, the system retrieves channels related to geometric positioning in the channel sequence, such as clamping posture channels, tooling reference surface channels, and contact confirmation channels, and extracts the stable intervals of these channels within the start and end frames of this segment as the original supporting data for geometric measurement point features. For frames with missing masks or weak sampling identifiers, the geometric measurement point features only record the positioning status and reference frame, without generating statistics, thus maintaining consistency with the principle of only marking and not estimating in the previous step. The second type of feature is dynamic features, defined as channel-derived quantities reflecting the dynamic response of the equipment and workpiece during processing. These include energy distribution in vibration channels, load changes in the spindle and feed, gradual changes in temperature channels, and the sequence of impact events. The system reads the channel list from the instance context and performs intra-segment statistical and morphological analysis on channels belonging to the dynamic response category. The statistical and morphological analysis follows the constraint of not introducing estimations; for missing intervals, only event indications are recorded without supplementary calculations. The third type of feature is motion control features, defined as the representation of control behaviors related to toolpath, feed commands, and attitude. These include path segment indications, acceleration / deceleration behaviors, attitude switching, and tool change trigger associations. The system extracts the control system version and readable fields of the path segment from the instance context, locates the control behaviors on the segment time axis, and maps them to anchor points in the alignment mapping table to form a sequence of control behaviors within the segment. After extracting the above three types of features, the system establishes a feature container for each operating mode segment. The feature container uses geometric measurement point features, dynamic features, and motion control features as three primary keys, storing corresponding feature subsets and indexes respectively.

[0036] The cross-domain feature alignment process is executed within the feature container, aiming to establish correspondences between the three types of features under the same temporal and spatial references. The first step in alignment is temporal alignment. The system calls the set of alignment anchor points left over from the previous section in the instance context, and performs frame-level mapping between event indicators in the dynamic features and path segments in the motion control features according to the anchor point time. When an event occurs within the area covered by a missing mask, only the adjacency relationship between the event and the path segment is recorded, without further inference of the temporal location. Spatial alignment is used to connect geometric measurement point features with motion control features. The system reads the target surface and measurement point positioning references from the process route list, compares the mapping of the measurement points in the workstation coordinates with the attitude switching in the path segment, and obtains a pairing table of measurement points and attitudes. To handle one-to-many or many-to-one relationships, the system records pairing weights in the pairing table. These weights do not participate in numerical calculations but are only used as sorting and filtering references. After completing the temporal and spatial alignments, the system enters the fusion generation stage, which aggregates the three types of features into fusion entries under the same segment and the same pairing key. Each fusion entry includes a measurement point identifier, target surface, path segment, attitude segment, dynamic event sequence, and environmental context. The environmental context is derived from the instance context and records a summary of the equipment model, control system version, and boundary identifier. For anomaly handling, when a segment lacks key support for any type of feature, the system still outputs a fusion entry, but marks the missing type in the entry for subsequent extraction of process feature elements to select a usable subset. The output of the above processing is a fusion feature set, which is organized in data structure by part number, process sequence, and segment index, and provides dual indexes for measurement point identifiers and path segments at the entry level for easy subsequent retrieval. This fusion feature set is directly consumed by the next subsection of this main step, serving as the sole source in the subsequent extraction of process feature elements and construction of element number and element-data mapping relationships, generating the process feature element set structure. Simultaneously, in cross-main step linkage, the fusion feature set also provides the binding basis for measurement points and attitudes for model component assembly in the virtual measurement stage, and provides a combined event and path reference for segment attribution in the error evaluation stage.

[0037] S230. Extract process feature elements from the fused feature set and construct the mapping relationship between element number and element-data to generate the process feature element set structure. Specifically, the fused feature set output in the previous section is used as input. This fused feature set has already established the temporal and spatial correspondences of geometric measurement point features, dynamic features, and motion control features at the fragment level. The object of the process feature element extraction is to describe stable and reusable process units in the machining process. The process feature element is defined as a structured description formed around the target surface, process segment, and tool action, including element type, element boundary, element association, and element supporting evidence. The element type is used to indicate the element's belonging in the process semantics, such as contour machining, hole system machining, end face machining, or profile following; the element boundary is used to define the fragment range, measurement point set, and path segment set covered by the element; the element association is used to record the association relationship between the element and equipment instance, tool number, and operation mode segment; the element supporting evidence is the set of fused entries and alignment anchor points extracted during the extraction process. The system first calls the process route list, compares the process sequence with the target surface, and bins the items in the fusion feature set according to the process. Within each process bin, the system constructs a candidate element chain based on the continuity of path segments and the stability of attitude segments. The starting point of the candidate element chain is the boundary or attitude switching point of the path segment, and the ending point is the next boundary or switching point. If a segment with impact or abnormal boundary markers appears in the dynamic event sequence, it is broken on both sides of the segment to form multiple candidate segments. Subsequently, the system scans the measurement point identifier set of each candidate segment. If there is a stable mapping between the set and the measurement point references in the process route list, the candidate segment is upgraded to a process feature element, and the element type and element boundary are written accordingly. If the measurement point identifier set is missing or not completely consistent with the process route list, the system can still generate the element, but a missing item label is added to the element supporting evidence. The label does not prevent subsequent use, but is only used as a reference for adjusting the sampling strategy during the virtual measurement stage.

[0038] During the element numbering process, the system employs a hierarchical numbering strategy, ensuring that elements have a unique and traceable identity across processes and segments. The hierarchical numbering consists of a part number layer, a process layer, an element type layer, and a segment sequence number layer. When generating numbers, the system reads relevant indexes from the fusion feature set and the process route list to ensure that similar elements of the same part maintain a stable order across different batches. When segment overlaps occur due to concurrent processing, the segment sequence number layer provides a serialized number based on the alignment anchor points' order on the timeline. After numbering, the system writes the number into the element's metadata and registers it in reverse on the corresponding entry in the fusion feature set, allowing the original entry to reference its associated element number, forming a bidirectional reference from entry to element and from element to entry. To prevent numbering conflicts, the system prepares a occupancy table. The occupancy table does not release numbers within the current batch until the model component assembly is completed in the virtual measurement phase, preventing the occurrence of identical numbers with different meanings during iterative processing within the batch. After numbering is completed, the system proceeds to construct the element-data mapping relationship, which clarifies the link between each element and its supporting data source. The mapping relationship comprises four types of fields: entry index, channel reference, instance context, and alignment anchor extract. The entry index is used to quickly return to the original entry in the fused feature set; the channel reference is used to locate the original channel that constitutes the evidence for that element; the instance context provides a summary of the device model, control system version, and boundary markers; and the alignment anchor extract allows the element to correspond to the sampling path of the virtual measurement on the timeline. Mapping relationship construction operates in both batch and incremental modes: batch processing generates mappings for the entire batch of fused feature sets at once; incremental processing updates corresponding elements or supplements mappings for newly added or revised fused entries. In exceptional circumstances, such as when an entry that an element depends on is marked as missing or weakly sampled, the system still generates a mapping, but adds a weak evidence marker to the mapping to facilitate subsequent judgment of evidence strength during model component assembly.

[0039] Through the extraction of process feature elements, the construction of element-data mapping relationships, and the establishment of process feature element set structures, the system generates a process feature element set structure. This structure serves as an organized input for the virtual measurement model, storing data content by grouping by part number and process sequence, layering by element type, and indexing by element number. Each element node contains five core fields: element type, element boundary, element association, supporting evidence, and element-data mapping relationship. This process feature element set structure is directly consumed by the next main step. In subsequent steps of acquiring process feature element sets, designing tolerance benchmarks and measurement rule sets, and assembling model components and binding rules, it becomes the sole element input participating in the assembly of model components in the virtual measurement model. Simultaneously, the element-data mapping relationships recorded in this structure provide a pathway from visualized data back to the original channel for the 3D chromatographic display and error assessment stages, supporting consistent referencing and traceability across steps. In summary, the technical effects of this step are as follows: by extracting and aligning cross-domain features from standardized instance data, and then precipitating them into structured process feature elements and bidirectional mapping relationships, a process-side input that can directly drive the assembly of the virtual measurement model is formed, and a traceable semantic anchor is provided for subsequent visualization and evaluation.

[0040] Step S300 includes at least steps S310-S330: S310. Obtain the set of process feature elements, design tolerance datum and measurement rule set, perform model component assembly and rule binding processing to obtain the virtual measurement model; Specifically, the process feature set output from the previous main step is used as the core input. This set records feature types, feature boundaries, feature relationships, supporting evidence, and feature-data mapping relationships. Feature types identify semantic categories such as contour machining, hole machining, end-face machining, or profile following. Feature boundaries define the segment range and measurement point set. Feature relationships indicate the binding relationship between equipment instances and tool numbers. Supporting evidence describes the formation chain from data to features. Feature-data mapping relationships provide entry indexes, channel references, and alignment anchor point extractions. Within the same input batch, design tolerance datums and measurement rule sets are loaded simultaneously. The design tolerance datum defines the tolerance type, tolerance zone direction, tolerance limits, and reference datum for each target surface. The measurement rule set defines the abstract specifications, sampling density, probe posture constraints, path constraints, contact criteria, and handling strategies for abnormal segments of the measuring tool. During the input loading phase, the system groups the process feature set according to part number and process sequence, and matches each group with the design tolerance datum entries. The matching principle is that the feature type is consistent with the target surface and the feature boundary covers the evaluation area of ​​the tolerance datum. If a feature meets the conditions with multiple tolerance datums, the primary datum is selected according to the priority selection entries in the measurement rule set, and the remaining entries are registered as backup datums. Backup datums do not participate in the assembly of this batch of models, but only serve as candidate configurations for subsequent switching. For cases marked as weak evidence in the feature-data mapping relationship, the label is retained during the input loading phase and passed to the confidence annotation of the model component during the subsequent assembly process, without changing the assembly order of the link. After completing the grouping and matching, the system enters the model component library for retrieval. The model component library abstracts measurement tools, sampling strategies, error evaluation operators, and visual binders into assemblable components, and records the input and output interfaces and rule dependencies for each component. The retrieval process determines the minimum coverage of the component set based on the feature type and design tolerance datum. If a feature lacks a dedicated component, a general component of the same type of feature is used, and a general substitute is noted in the confidence annotation.

[0041] During the component assembly stage, the system constructs a local workspace for simulation measurement based on the feature boundaries of the process feature set. This local workspace only covers the target surface and its adjacent necessary areas, and is bound to corresponding equipment instances, tool numbers, and operating mode segments through feature association. Subsequently, a sampling strategy and probe attitude constraints are selected for each feature according to the measurement rule set. The sampling strategy determines the distribution and density of sampling points on the target surface, while the probe attitude constraints limit the incident direction and offset range of the probe or beam relative to the target surface. Path constraints are used to generate feasible sampling paths, referencing alignment anchor points in the feature-data mapping relationship. Temporal and spatial associations limit the synchronization relationship between the sampling path and the segment time axis and equipment attitude switching. Contact criteria provide triggering conditions and stability determination for contact-type simulation tools, while non-contact simulation tools here transform into a composite determination of echo quality and distance threshold. The rule binding process follows the sequence of baseline selection—tool assembly—sampling strategy activation—attitude constraint activation—path constraint activation—criteria activation. Once a rule binding takes effect, the version and timestamp are fixed on the component instance, and the cross-reference between the process feature element number and the design tolerance baseline number is retained in the instance metadata, forming a traceable binding pair. For feature boundaries with abnormal segments, rule binding switches the sampling strategy within the abnormal segment to a read-only strategy. The read-only strategy does not dispatch new sampling tasks, but only records the event location and uses it as a masking or de-weighting condition during the evaluation phase. After completing component-level rule binding, the system enters model framework assembly. The model framework performs topological organization on each feature component within the current process bucket. Common organization methods include parallel measurement under the same target surface, serial measurement subject to sequence dependence, and conditional measurement across features. Topological organization explicitly records the front-to-back constraints and synchronization points between components in the assembly drawing. Front-to-back constraints come from sampling path conflicts and mutual exclusion of equipment instances, while synchronization points come from alignment anchor points and operation mode segment boundaries. If a mutually exclusive resource conflict is discovered during the organization process and cannot be resolved within the current window, the system will register the conflict pair in a delayed queue and apply a degradation strategy to the relevant element components in this batch of models. The degradation strategy is either a reduction in sampling density or a reduction in rendering priority, and the reason for degradation and the conditions for resolving it will be recorded in the metadata. After the model framework is assembled, the system generates an instantiated object of the virtual measurement model. The instantiated object contains an assembly drawing, rule binding records, a list of component instances, and a local workspace description. This object is the output product of this subsection, with the output field named "Virtual Measurement Model." It will be directly consumed by the next subsection of this main step. In subsequent steps, the target surface area, sampling path, and attitude parameters will be extracted from the virtual measurement model for simulation measurement and result aggregation, serving as the sole input in the generated virtual measurement results. Simultaneously, in the cross-main step linkage, the virtual measurement model will also be referenced by its assembly drawing and local workspace description in the subsequent 3D chromatography display data generation and error evaluation stages for spatial registration and region clustering.

[0042] S320. Extract the target surface area, sampling path, and attitude parameters from the virtual measurement model, perform simulation measurement and result aggregation, and generate virtual measurement results. Specifically, the aforementioned virtual measurement model serves as the entry point for execution. This virtual measurement model has already embedded assembly drawings, component instances, and rule binding records. During the preparation phase, the system reads the topological order from the assembly drawing, generates an execution sequence according to a combination strategy of parallel and serial measurement, and extracts the target surface region from the local workspace description. The target surface region is used to define the spatial range of the simulation measurement, including definitions of target surface boundaries, adjacent transition zones, and safety clearance zones. On the same component element, the system extracts sampling paths and attitude parameters based on the rule binding records. The sampling path is a discrete sequence of feasible paths within the target surface region, and the attitude parameters are the detection attitude or sensing attitude when executing along the path. Together, they constitute the motion and observation configuration of the simulation measurement. To adapt to the physical limitations of the equipment instance, the preparation phase also aligns the path constraints with the equipment travel, acceleration / deceleration capabilities, and tool change timing. If the travel or synchronization requirements are not met, the path is segmented and rearranged, and the rearrangement record is written into the execution sequence annotation. After preparation, the system initiates simulation measurement. The execution engine schedules component instances according to the execution sequence. For contact components, it performs probe contact, stability determination, and sampling point recording; for non-contact components, it performs projection and echo acquisition, echo validity determination, and sampling point recording. All sampling is performed under criteria fixed during assembly. If a criterion is not triggered or is unstable, the execution engine does not generate a replacement value, but only writes an invalid flag at the sampling point and accumulates an invalid count in the instance context to facilitate segment masking during result aggregation. For parallel measurement branches in the topology organization, the execution engine merges them at the synchronization point set. The merging behavior depends on the alignment anchor point and the boundary of the running mode segment. If a branch enters a read-only policy due to an abnormal segment, that branch only provides a time position reference at the merging point and does not participate in sample value aggregation.

[0043] In the result aggregation phase, the system collects sampling points and corresponding attitudes for each element component, forming component-level measurement segments. Subsequently, at the element level, these segments are spatially reordered and merged. Reordering follows the natural traversal order of the target surface boundary and the principle of stability priority in attitude switching. Segment merging is performed under conditions of boundary continuity and consistent criteria. For cross-element conditional measurements, result aggregation establishes a reference relationship between the sampling segments of subsequent elements and the reference segments of preceding elements based on the constraints in the assembly drawing, enabling comparison of conditional associations in subsequent evaluation stages. During segment merging, sampling points with invalid markers are retained as event points, not participating in numerical statistics but retaining their spatial and temporal positions. The sparsity or density of sampling points is controlled by the sampling strategy, without triggering automatic interpolation or deletion; only segments with density deviations are recorded in the metadata. After aggregation, the system generates element-level measurement results for each element, including a set of sampling points, attitude sequences, invalid event points, segment merging records, and cross-references in the assembly drawing. Furthermore, within the process bucket, the system reassembles the measurement results at each element level according to the target surface to generate process-level measurement results. These process-level measurement results retain element boundaries and record the relative positions of overlapping samples within common areas, avoiding unclear rendering order during the visualization stage. For elements that are downgraded due to resource conflicts, the system retains downgrade labels in the results, allowing for the selection of appropriate texture precision or layering strategies during spatial registration and color coding in the next subsection. Thus, the output of this subsection is the virtual measurement result, with the output field named "Virtual Measurement Result." This result is directly consumed by the next subsection and serves as the sole source for subsequent spatial registration, color coding, and layered rendering of the virtual measurement result, generating 3D chromatographic display data. Simultaneously, in the cross-main-step data link, the virtual measurement result provides the input form for geometric deviation comparison in the error evaluation stage and provides the raw data for spatial registration in the 3D chromatographic display.

[0044] S330. Perform spatial registration, color mark encoding, and layered rendering on the virtual measurement results to generate three-dimensional chromatographic display data; Specifically, the aforementioned virtual measurement results are used as input. These virtual measurement results include element-level and process-level measurement results, recording the sampling point set, posture sequence, invalid event points, and fragment merging records. Spatial registration processing first reads the local workspace description and alignment anchor point extraction written during assembly of the process feature element set, and extracts reference datum and orientation information from the design tolerance datum to construct a registration mapping from the local workspace to the part reference coordinates. The registration steps are performed at the element level and process level respectively: element-level registration projects the sampling point set onto the evaluation area of ​​the target surface, binds the corresponding element boundaries, and corrects the ambiguity of the attribution of the points at the boundaries; process-level registration processes the common areas between elements, determines the visualization priority order of the common areas based on the overlapping annotations in the fragment merging records and the front and back constraints of the assembly drawing, and generates the transition zone of the splicing gap. For fragments containing invalid event points, spatial registration only retains their spatial position and does not generate visualization intensity values ​​to avoid misjudgment in subsequent color code encoding. After spatial registration, the system proceeds to color mark encoding. The goal of color mark encoding is to represent the virtual measurement results on the target surface as a color distribution with intuitive gradients and discriminative power. Before encoding, the system selects an appropriate difference measure from the virtual measurement results based on the tolerance type and limits of the design tolerance benchmark, and constructs the input domain for color mark mapping. During encoding, the system calculates the location value within the input domain for each sampling point and maps the location value to color according to the color mark scheme specified in the measurement rule set. Common color mark schemes include monotonic gradient color marks and segmented threshold color marks. To handle color mark conflicts in common areas, the system employs a layered strategy in the splicing transition zone, superimposing colors from different elements within the transition zone according to priority and transparency. The transparency strategy is derived from the parallel measurement weights or degradation annotations in the assembly drawing. For degradation elements, color mark encoding uses a low-resolution texture or a scheme emphasizing texture edges, and the degradation reason is written in the texture metadata for reference in the subsequent error assessment stage of region clustering.

[0045] Layered rendering is performed after color mark encoding, aiming to construct a multi-layered visible color spectrum representation in 3D space. Layered rendering first generates a rendering stack based on the hierarchical relationship of the target surface and feature boundaries. The bottom of the stack contains the base geometry, the middle contains the feature color layers, and the top contains the event and annotation layers. During rendering, the system traverses each layer along the execution sequence, attaching color mark textures to the corresponding target surface areas. Common areas are overlaid using the aforementioned transparency strategy. For invalid event points, the system presents them as markers in the annotation layer, carrying fragment and assembly references for easy interactive retrieval. To adapt to heterogeneous display terminals, layered rendering generates multi-resolution versions during the output stage and registers the coordinate system, layering order, and texture mapping rules in the metadata. The final output 3D color spectrum display data contains four core categories: registration geometry, color mark textures, layer descriptions, and cross-references, which can be directly read by subsequent analysis steps. This output, as a product of this subsection, has the field name "3D Chromatography Display Data." It is used in the next main step to perform geometric deviation comparison and region clustering processing based on the 3D chromatography display data, resulting in an error label sequence that is directly consumed. Simultaneously, in the cross-main step linkage, the 3D chromatography display data also provides spatial region retrieval and visualization indexes for error assessment and processing status prediction stages, and provides the location basis for target surfaces and feature boundaries for generating parameter correction instructions during reinjection. In summary, the technical effects of this step are: by completing spatial registration of sampling points, color coding of difference measurements, and layered rendering based on target surfaces under the drive of a virtual measurement model, 3D chromatography display data consistent with process elements and tolerance benchmarks is formed, and a visualization data channel that can be directly called upon by subsequent geometric deviation comparison and region clustering is established.

[0046] Step S400 includes at least steps S410-S430: S410. Based on the three-dimensional chromatographic display data, perform geometric deviation comparison and region clustering to obtain the error label sequence; Specifically, the 3D chromatographic display data output from the previous section is used as the sole input. This 3D chromatographic display data includes four core categories: registration geometry, color mark texture, hierarchical description, and cross-references. It establishes reference relationships with the process feature element set, assembly drawing, and alignment anchor points at the metadata level. During the input loading phase, the system first reads the target surface and common area annotations in the registration geometry, reads the element color layer, event, and annotation layer in the hierarchical description, and parses the cross-references to obtain the corresponding entries of the sampling points on the time axis and process element numbers. For areas with downgraded annotations and invalid event points, the system adds weak evidence and read-only segment identifiers to the loading table. These areas only participate in geometric positioning and cluster boundary constraints, not numerical aggregation. After input loading is completed, the system enters the geometric deviation comparison processing. The geometric deviation comparison is defined as establishing a difference metric between the design tolerance benchmark reference and the color mark texture encoding, and forming a deviation distribution description within the spatial neighborhood. The processing flow begins on the target surface, traversing block by block according to the element boundaries in the layered description. The positioning values ​​of the color mark texture on the target surface are compared with the limits of the design tolerance datum. The comparison results, along with the attitude sequence of the sampling points, the front and rear constraints in the assembly drawing, and the visualization priority of common areas, are written into the deviation fragment cache. For splicing gaps and transition zones, the system establishes overlay records in the deviation fragment cache according to the parallel measurement weights and transparency strategy of the assembly drawing, preventing duplicate entries. When multiple elements overlap on the same surface, the deviation fragment cache retains an independent record for each element and establishes cross-references in common areas for use in the region clustering stage. In abnormal situations, such as when a fragment contains only invalid event points, the system only registers the spatial and temporal locations in the deviation fragment cache, does not generate deviation positioning values, and adds a non-aggregation identifier to the fragment header.

[0047] In the region clustering process, the system performs spatial clustering and semantic aggregation on the deviation fragment cache according to a two-layer organizational order of target surface and process bucket. Spatial clustering uses registration geometry as a reference, grouping within the evaluation area of ​​the target surface according to neighborhood structure and curvature continuity. When encountering common areas, the cluster boundaries prioritize the priority order and transition band width in the hierarchical description, generating fine-grained sub-clusters within the transition band when necessary for subsequent error label subdivision. Semantic aggregation uses process feature element numbers and path segments as clues, merging spatially adjacent deviation fragments with consistent sources into candidate regions, and writing the relative positions of alignment anchor points on the time axis into the time annotations of the candidate regions. The time annotations provide anchors for establishing correspondences with equipment operation segments and tool usage segments. For regions labeled as weak evidence and read-only segments, semantic aggregation only retains their geometric range and source index, without cross-region merging. After completing spatial clustering and semantic aggregation, the system generates region entries, each bound to the target surface, feature number, path segment, time annotation, overlay record, and downgrade label, and sends the region entries to the label generation unit. The label generation unit assigns label names, label boundaries, and label references to region entries based on the tolerance type, tolerance zone direction, and limits of the design tolerance datum. Label names are used for quick retrieval of similar regions in subsequent steps; label boundaries describe the closed contour on the target surface and the start and end segments on the time axis; and label references are used to trace back color mark textures and sampling point sets. To avoid overlapping labels caused by common areas, the label generation unit performs deduplication and splitting on region entries with overlapping records: it maintains primary priority labels at overlapping points and generates subordinate labels for secondary priority labels. Subordinate labels only serve a positioning function in subsequent evaluation. At this point, the output of this subsection is the error label sequence, which is composed of the region entry sequence and label meta-information, with the field name "Error Label Sequence". This error label sequence serves as the direct input to the next subsection of this main step. It is used to extract feature segments and corresponding equipment operation segments and tool usage segments from the error label sequence, perform error assessment and equipment health status mapping, and generate error assessment results. At the same time, in cross-main step association, the error label sequence can be used to look up the 3D chromatographic display data and virtual measurement results through its label references, and when necessary, it can be associated with the element number and element-data mapping relationship in the process feature element set.

[0048] S420. Extract feature segments and corresponding equipment operation segments and tool usage segments from the error label sequence, perform error assessment and equipment health status mapping, and generate error assessment results. Specifically, the aforementioned error label sequence is used as input, and this sequence already includes region entries, label metadata, label boundaries, and label references. During the preparation phase, the system accesses the 3D chromatographic display data and virtual measurement results through label references, locates the sampling point set and attitude sequence corresponding to each label, and further traces back to the standardized instance data and aligned segmented data structure through the element-data mapping relationship in the process feature element set to obtain the operating mode segment index within the label's coverage time range. Based on this index, the system extracts the time range matching the label boundary from the aligned segmented data structure to form feature segments; simultaneously, based on the equipment instance context and the tool list in the process route list, it retrieves equipment operation segments and tool usage segments within the same time range. The feature segment is defined as a combination of multi-channel observations and attitude records covered by the error label boundary; the equipment operation segment is defined as a sequence of equipment control states and load states within the same time range; and the tool usage segment is defined as a combination of tool number, tool position life, and tool change record. In abnormal situations, such as when the time annotation of a certain tag does not overlap with the running mode segment, the system will classify the tag as a cross-segment tag. Cross-segment tags will still generate feature segments, but they will only be used as reference items in the evaluation and will not participate in the rule triggering count.

[0049] After segment extraction is completed, the system enters the error evaluation process. This error evaluation is based on label type, observed features of the feature segments, the state of the equipment operation segment, and the lifespan stage of the tool usage segment. Triggers and attributions are performed according to evaluation entries in the rule base. The rule base consists of basic evaluation entries and combined evaluation entries: basic evaluation entries are used to identify the causes of deviations from a single source, such as geometric errors related to attitude or load anomalies related to feed; combined evaluation entries are used to synthesize evidence from multiple sources to provide composite attributions, such as identifying thermally induced deformation when path segment transitions and temperature gradients occur simultaneously. During evaluation execution, the system iterates through the error label sequence one by one, inputting the feature segments, equipment operation segments, and tool usage segments corresponding to each label into the evaluation engine. The evaluation engine performs segment comparison and semantic matching of the observed features and state sequences without modifying the original values ​​or interpolating missing values. When all trigger conditions of the rule entries are met, an attribution record is generated and attached to the corresponding label. The attribution record includes an attribution name, triggering evidence index, fragment cross-references, and confidence notes. The triggering evidence index references the observed feature location within a feature fragment, the state transition point within an equipment operation fragment, and the lifespan stage location within a tool usage fragment. Fragment cross-references are used to trace back to alignment anchors and assembly drawings. Confidence notes are retained in the weak evidence read-only segments and downgraded annotations passed upstream. When the same label satisfies multiple evaluation criteria, the system makes a decision based on the priority and mutual exclusion relationships in the rule base. When there are parallel and non-exclusive criteria, the evaluation engine generates multiple attributions for that label and records the priority order in the label metadata for subsequent trend analysis and working condition matching. If the evaluation engine does not find a matching criterion, the system outputs a non-attributed record and pushes it to the rule audit queue for expanding evaluation criteria in subsequent batches. Through the above evaluation and attribution process, the system forms a set of evaluation criteria for each label and summarizes them into a batch-level evaluation structure at the process bucket dimension. The final output is the error assessment result, which includes label-attribution pairs, fragment cross-references, priority sorting, and audit queue indexes. The field name is "Error Assessment Result." This error assessment result is directly consumed by the next subsection of this main step and serves as the sole input in the subsequent generation of trend analysis, operating condition matching, and parameter correction instructions for the error assessment result, as well as in the model update instruction structure. Simultaneously, in the collaboration across main steps, the error assessment result establishes a backtracking path through fragment cross-references, 3D chromatographic display data, virtual measurement results, and process feature element sets.

[0050] S430, Perform trend analysis, working condition matching, and parameter correction instructions on the error evaluation results, and generate the model update instruction structure; Specifically, the aforementioned error assessment results are used as input, and these results aggregate label-attribution pairs, fragment cross-references, and priority ranking. The goal of trend analysis is to identify recurring attribution patterns and regional hotspots from both time and batch dimensions. During the preparation phase, the system extracts alignment anchor time and process sequence based on fragment cross-references to construct a batch timeline; simultaneously, it establishes a spatial grid based on target surfaces and element numbers to represent the frequency of label-attribution pairs in space. Trend analysis is performed in parallel on the batch timeline and the spatial grid: on the timeline, the system aggregates the occurrence locations of similar attributions in segments, recording three forms: continuous occurrence, intermittent occurrence, and phased occurrence; on the spatial grid, the system calculates the occurrence density of similar attributions within grid cells and associates the degradation labels in the hierarchical description with the coverage records of common areas, marking areas that may be affected by resource conflicts or rendering strategies. For unattributed records with audit queue indexes, trend analysis only describes the frequency in the time and spatial dimensions and does not participate in condition matching. After completing the trend analysis, the system writes the aggregated results of time and space into a trend summary. The trend summary and priority ranking together serve as the prerequisite for working condition matching.

[0051] In the condition matching process, the system reads the list of operating mode segments from the aligned segmented data structure and the instance context from the standardized instance data, and imports the condition definitions and allowed intervals from the process route list. The task of condition matching is to match the attribution patterns in the trend summary with the actual operating modes, equipment states, and tool stages to locate the condition combinations that may lead to the attribution pattern. The matching process is executed category by category: for attitude-related attributions, the system prioritizes searching for condition combinations in the neighborhood of path segments and attitude transitions; for load-related attributions, the system prioritizes searching for condition combinations in the spindle and feed state segments; for tool wear-related attributions, the system prioritizes searching for condition combinations in the tool's lifespan segment. During matching, the system follows the principles of no interpolation and no rewriting, only marking missing intervals and reducing the confidence note of the combination in the matching results. If a certain attribution pattern has supporting evidence across multiple operating condition combinations, the system assigns a matching order based on the priority ranking in the error assessment results and the process importance in the process route list. When the matching result is empty, the system adds the pattern to the list of combinations to be expanded for subsequent updates to the operating condition definition. After completing the operating condition matching, the system proceeds to generate parameter correction instructions. These instructions are defined as executable configuration updates for the virtual measurement model, 3D chromatography display, and data acquisition link, including fine-tuning of the time base configuration, rearranging of sampling path segments, adjusting the range of attitude constraints, switching sampling density in specific regions, and starting / stopping rule base entries. Instruction generation is based on the spatial hotspots of the matched operating condition combinations and trend summaries, corresponding to the assembly drawing nodes and rule binding records of the virtual measurement model. For example, if repeated attributions related to attitude appear in the hotspot area of ​​a specific target surface, the instruction generation outputs the range adjustment of attitude constraints and the refinement of the sampling path on that element component. If a phased attribution related to load appears within a specific operating mode segment, the instruction generation outputs the fine-tuning of the time base configuration and the revision of the start and end of the sampling window in the data acquisition link. All commands are generated with cross-references to fragments and alignment anchors to facilitate precise integration with existing configurations during back-injection. For scenarios with weak evidence, command generation uses a read-only strategy to output observational commands, recording only the location and enabling monitoring without modifying existing configurations.

[0052] The final output is a model update instruction structure, which consists of four types of fields: configuration update items, target node references, trigger conditions, and rollback conditions. The field name is "model update instruction." This model update instruction is injected back into the aforementioned process of acquiring part status data, equipment status data, and sensor timestamps from the production line. It performs unified time reference configuration and drift correction processing to obtain the configuration management module of the perception dataset. This module executes and takes effect at the sampling boundary or scheduling window. At the same time, this model update instruction can be applied to component instances and rule binding records of the virtual measurement model through target node references. The next batch extracts the target surface area, sampling path, and attitude parameters from the virtual measurement model, performs simulation measurement operation and result aggregation, and generates virtual measurement results for automatic consumption. This forms a closed loop of evaluation-update-remeasurement on the data channel across the main steps. In summary, the technical effects of this step are as follows: By performing trend analysis in time and space dimensions based on the error evaluation results, and combining the operating mode and equipment status for condition matching, data and model configuration update items that can be reinjected are generated; the model update instruction structure forms a consistent executable modification entry point in the acquisition, modeling and display links, supporting continuous revision and convergence in subsequent batches.

Claims

1. A virtual measurement and simulation method for processing quality based on ubiquitous sensing, characterized in that, include: Acquire part status data, equipment status data and sensor timestamps from the processing production line, perform unified time reference configuration, drift correction, channel denoising and step size resampling, mark missing intervals without estimating, cross-device clock alignment and event window truncation and operation mode label segmentation, and generate an aligned segmented data structure organized by segments; The aligned segmented data structure is obtained, and channel standardization, geometric measurement points, dynamics, motion control feature extraction and cross-domain alignment fusion, feature numbering and feature-data mapping processing are performed to generate a process feature feature set structure. Based on the structure of the process feature set, virtual measurement model assembly, simulation measurement based on target surface area, sampling path and attitude parameters, spatial registration, color mark encoding and layered rendering are performed to generate three-dimensional color display data. Based on the three-dimensional chromatographic display data, geometric deviation comparison and region clustering, error evaluation and equipment health status mapping and trend analysis and working condition matching are performed to generate parameter correction instructions and obtain the model update instruction structure.

2. The method according to claim 1, characterized in that, The part status data, equipment status data, and sensor timestamps from the production line include: The part status data of the processing line includes the process progress of the part at the workstation, the clamping posture and number association, as well as design tolerance references and measurement point location references; Equipment status data includes machine tool spindle speed, feed rate, tool number, servo load, temperature and vibration monitoring readings, and records its data identifier, unit, sampling period, trigger condition and abnormal code in the channel description to support cross-equipment standardization and operation mode label segmentation; Sensor timestamps are generated by edge acquisition nodes at the moment of each sampling and appended to the original measurement value, subject to a unified time reference configuration.

3. The method according to claim 1, characterized in that, The process of generating an aligned segmented data structure organized by fragments also includes: Acquire part status data, equipment status data and sensor timestamps from the processing production line, establish channel descriptions through the fieldbus gateway and data access module, load NTP and PTP dual-stack alignment and drift correction with unified time reference configuration, and only mark time anomalies without rewriting them to generate a perception dataset organized by workstation, equipment and part number. The channel list, sampling step size and missing segment label are extracted from the perception dataset. Denoising is performed by selecting a rule base strategy based on channel type, resampling at the target time axis step size and only labeling the missing interval without estimating the value, and a structured time series with a three-level index and a missing segment dictionary is generated. The alignment mapping table anchors are obtained from the structured time series. Cross-device clock alignment, event window truncation based on joint thresholds and anchors, and tag segmentation processing according to the operation mode dictionary are performed to generate an aligned segmented data structure containing an event window list and an operation mode fragment list.

4. The method according to claim 1, characterized in that, The process of generating the structure of the process feature set also includes: Based on the aligned segmented data structure, channel standardization is performed with unified units, unidirectional direction and consistent boundaries. Equipment instance mapping is then performed according to the rules of equipment number, channel source and workstation binding to obtain standardized instance data organized by operation mode segments. Geometric measurement point features, dynamic features, and motion control features are extracted from standardized instance data. Temporal alignment based on alignment anchor points and spatial alignment based on target surface and measurement point references are performed. Cross-domain fusion processing is performed by aggregating segments to generate a fusion feature set containing dual indices of measurement points and path segments. The feature set is used to obtain the feature type, boundary, association and supporting evidence. Candidate feature chains are constructed by binning by process and path continuity, and layer numbering and feature-data mapping relationship are constructed to generate the process feature feature set structure.

5. The method according to claim 1, characterized in that, The process of generating three-dimensional chromatographic display data also includes: Based on the structure of the process feature element set, the model component library is retrieved, the local workspace is constructed, the sampling strategy is implemented, the rule binding of posture and path constraints is performed, and the assembly drawing topology is processed to obtain the virtual measurement model. The target surface area, sampling path and attitude parameters are extracted from the virtual measurement model, and the simulation measurement operation and element and process level result aggregation processing are performed according to the assembly drawing execution sequence to generate virtual measurement results. The registration geometry and fragment merging records are obtained from the virtual measurement results. The spatial registration is driven by the reference datum, the color code is selected according to the tolerance limit to select the difference measure, and the layered rendering processing based on the feature color layer and annotation layer is performed to generate three-dimensional color display data.

6. The method according to claim 1, characterized in that, The process of obtaining the model update instruction structure also includes: Based on 3D chromatographic display data, geometric deviation comparison under design tolerance datum and assembly drawing constraints, disambiguation of common area coverage records, and spatial clustering by curvature and neighborhood, combined with semantic aggregation processing of element number and path segment, are performed to obtain error label sequence. Feature segments, equipment operation segments, and tool usage segments that match the label boundaries are extracted from the error label sequence. Error judgment and equipment health status mapping are performed based on the basic and combined entries of the rule base to generate error judgment results. The label-attribution pairs and fragment cross-references are obtained from the error evaluation results. Trend analysis of batch timelines and spatial grids is performed, and working conditions are matched according to operating mode, equipment status, and tool stage. The process of re-injectable time base fine-tuning, sampling path refinement, and attitude constraint adjustment update items is output to generate the model update instruction structure.

7. The method according to claim 5, characterized in that, The process of running the virtual measurement model also includes: After reading the topological order in the assembly drawing, the target surface region is extracted from the local workspace description. The target surface region is used to define the spatial range of the simulation measurement, including the target surface boundary, the adjacent transition area, and the safety clearance area.

8. The method according to claim 7, characterized in that, On the same element component, sampling paths and attitude parameters are extracted. The sampling path is a discrete sequence of feasible paths within the target surface area, and the attitude parameters are the detection attitude or sensing attitude when executing along the path. Together, they constitute the motion and observation configuration of the simulation measurement.

9. The method according to claim 8, characterized in that, During the preparation phase, path constraints are aligned with equipment travel, acceleration / deceleration capabilities, and tool change timing. When travel or synchronization requirements are not met, the path is split into segments and rearranged in order, and the rearrangement record is written into the execution sequence notes.

10. The method according to claim 9, characterized in that, The sampling points and corresponding attitudes are collected to form component-level measurement segments. At the element level, the segments are spatially reordered and merged. The reordering follows the natural traversal order of the target surface boundary and the stability priority principle of attitude switching. The segment merging is carried out under the condition that the boundary is continuous and the criteria are consistent.

Images