A manufacturing equipment service state intelligent monitoring system and method

Abstract

The application belongs to the technical field of intelligent manufacturing, and provides a manufacturing equipment service state intelligent monitoring system and method, which comprises a data sensing module for collecting multi-source heterogeneous signals in the service process of manufacturing equipment; an edge computing module for performing process semantic analysis, equipment digital twin model credibility evaluation, adaptive monitoring scheduling, equipment baseline response normalization and online self-verification; a cloud module for storing event data and updating intelligent inference models and equipment digital twin models; and an on-machine self-verification benchmark module for providing online reference measurement when the model credibility decreases or the abnormal risk increases. The method dynamically schedules the sampling, conditioning, caching and storage processes according to the process semantic state, model credibility and abnormal risk, realizes the monitoring of the states of key components of equipment, process states and manufacturing objects, and improves the key transient information capture capability, monitoring result credibility and model cross-equipment migration capability.

Description

Technical Field

[0001] This invention belongs to the field of intelligent manufacturing technology, specifically relating to an intelligent monitoring system and method for the service status of manufacturing equipment, which is particularly suitable for online monitoring of the service status of manufacturing equipment such as ultra-precision machine tools, CNC machining equipment, grinding equipment, and polishing equipment. Background Technology

[0002] Manufacturing equipment is a crucial platform for material removal, surface shaping, precision finishing, and the fabrication of complex functional surfaces. Its service condition directly impacts the processing quality, dimensional accuracy, surface quality, and operational stability of the manufactured objects. Ultra-precision machine tools, CNC machining equipment, grinding equipment, and polishing equipment are typical examples of machining or surface treatment manufacturing equipment. During service, they are typically affected by a combination of factors, including the condition of process-executing components, thermal deformation, micro-vibration, tool or grinding wheel wear, process force fluctuations, surface contour errors of the manufactured objects, and environmental disturbances. As manufacturing processes evolve towards higher precision, higher stability, and continuous operation, relying solely on post-processing inspections or manual experience to judge equipment status is no longer sufficient to promptly reflect subtle changes during equipment operation. Therefore, utilizing multi-source sensors for online monitoring of the service condition of manufacturing equipment, combined with intelligent models for status identification, anomaly detection, and predictive analysis, has become an important means to improve manufacturing quality and equipment reliability.

[0003] Existing manufacturing equipment condition monitoring technologies typically acquire multi-source signals using sensors such as temperature sensors, vibration sensors, force sensors, acoustic emission sensors, displacement sensors, speed sensors, current sensors, and power sensors. These signals are then processed, stored, and analyzed by a host computer or edge computing device. However, existing technologies still have the following shortcomings: First, multi-source signals are often simply aligned using system time, lacking a correspondence with phase information of process execution components, position and attitude information of equipment motion execution mechanisms, process semantic state, and manufacturing object area number, making it difficult for the acquired data to accurately reflect specific process semantics. Second, sampling rate, range, gain, buffering mode, and storage strategy are mostly pre-fixed. Using the same acquisition method in different process semantic states such as standby, approach, process action establishment, steady-state operation, process action exit, rollback, and online detection easily generates a large amount of low-value data and may miss important information during critical transient phases. Third, although digital twin models can... It is used to predict the operating status of manufacturing equipment, but when sensor drift, equipment thermal state changes, clamping state changes, process load changes, or model parameters deviate, the model prediction results may deviate from the actual equipment state. Existing systems have difficulty evaluating the credibility of digital twin models and judging the source of deviation in a timely manner, resulting in a lack of credible basis for abnormal alarms, self-verification triggers, and model updates. Fourth, there are differences between different manufacturing equipment in terms of temperature rise baseline, vibration transmission characteristics, phase lag of process execution components, geometric error baseline, load response baseline, and energy input response baseline. When the intelligent model or digital twin model on a single piece of equipment is directly transferred to other equipment, it is easy to reduce the accuracy of state recognition, abnormal judgment, and predictive analysis.

[0004] To address the aforementioned shortcomings, there is an urgent need to develop an intelligent monitoring system and method for the service status of manufacturing equipment, in order to solve the problems of unclear process semantics, large amounts of low-value data, uncertain biases in digital twin models, and difficulties in transferring digital twin models. Summary of the Invention

[0005] This invention addresses the problems in current equipment service status monitoring processes, such as multi-source heterogeneous signals being synchronized only by time, fixed sampling modes, difficulty in dynamically adjusting the granularity of event data storage, difficulty in assessing the reliability of digital twin models, and insufficient cross-equipment migration capability of intelligent models. It provides an intelligent monitoring system and method for the service status of manufacturing equipment.

[0006] This invention discloses an intelligent monitoring system for the service status of manufacturing equipment, the intelligent monitoring system for the service status of manufacturing equipment comprising:

[0007] The data sensing module (100) is used to acquire multi-source heterogeneous signals generated during the service of manufacturing equipment;

[0008] The edge computing module (200) runs multi-source signal monitoring software (210), which includes:

[0009] The equipment digital twin model (212) is used to generate predicted service conditions for the current and future preset time windows of the manufacturing equipment;

[0010] The process semantic parsing submodule (213) is used to divide the service process of manufacturing equipment into different process semantic states based on the predicted service conditions, the position and attitude information of the equipment motion actuator, the phase information of the process execution component and the multi-source heterogeneous signal.

[0011] The digital twin model credibility assessment submodule (214) is used to calculate the credibility of the digital twin model based on the predicted service conditions;

[0012] The intelligent reasoning model (211) is used to output anomaly risks based on the credibility of the digital twin model;

[0013] The adaptive monitoring and scheduling submodule (215) is used to dynamically schedule the storage granularity based on the process semantic state, the credibility of the digital twin model, and the risk of anomalies.

[0014] The online self-verification submodule (216) is used to perform online reference measurement when the credibility of the digital twin model is lower than a preset threshold or the abnormal risk is higher than a preset threshold, and to correct the sensor and model parameters based on the measurement results.

[0015] The equipment baseline response normalization submodule (217) is used to establish the equipment baseline response model and perform baseline response normalization processing on the multi-source heterogeneous signals acquired online.

[0016] The cloud module (300) is used to store monitoring parameters and update the edge-side model parameters, and then send the updated model parameters to the edge computing module (200).

[0017] The on-machine self-verification benchmark module (400) is used to perform online verification when the edge computing module (200) issues a self-verification command.

[0018] Furthermore, the aforementioned abnormal risks Represented as:

[0019] ;

[0020] in, This represents the probability of an anomaly. For the abrupt change of multi-source signals, The risk weight corresponding to the current process semantic state. To enhance the credibility of digital twin models, For the Sigmoid function, These are the weighting coefficients.

[0021] Furthermore, the feature vector used for process semantic judgment is:

[0022] ;

[0023] in, To equip the digital twin model 212 with the predicted proximity distance, contact state, or operational margin between the process execution components and the manufactured object, This refers to the feed rate or relative motion speed. Spindle speed or speed of the process execution component. For the phase information of the process execution components, To provide position and attitude information for the motion actuators, Characteristics of force signals or load signals. Characteristics of acoustic emission, vibration, current, power, or other technological states;

[0024] For the set of process semantic states:

[0025] ;

[0026] The probability of the current time belonging to the i-th time step is calculated using the following probabilistic form. The probability of each process semantic state:

[0027] ;

[0028] in, This is the process semantic state number for which the probability is currently to be calculated. and The first The weight vector and bias corresponding to each process semantic state; This represents the total number of process semantic states. For summing indexes of process semantic states, and The first The weight vector and bias corresponding to each process semantic state;

[0029] The current process semantic state is:

[0030] ;

[0031] This represents the current technological semantic state.

[0032] Furthermore, the credibility of digital twin models Defined as:

[0033] ;

[0034] in, and This is the adjustment coefficient;

[0035] The residual between predicted service conditions and measured results of multi-source heterogeneous signals Represented as:

[0036] ;

[0037] in, To equip the digital twin model 212 at time The output predicted feature vector, The measured feature vectors of multi-source heterogeneous signals are... This is the residual weight matrix;

[0038] Let the first The health status of each sensor is Its weight is Then the sensor health status Represented as:

[0039] ;

[0040] in, Let be the number of sensors participating in the monitoring; let the number of channels in the current event data frame that have completed valid spatiotemporal marking be . The total number of channels that should complete spatiotemporal marking is Then the spatiotemporal marking integrity Represented as:

[0041] ;

[0042] The equipment baseline response offset is used to characterize the degree of deviation of the current equipment baseline response relative to the pre-established equipment baseline response model; let the current baseline response characteristics be... The mean vector of the equipment baseline response model is The covariance matrix is Then the equipment baseline response offset Represented as:

[0043] ;

[0044] The larger the prediction residual and the larger the equipment baseline response offset, the lower the credibility of the digital twin model; the better the sensor health status and the more complete the spatiotemporal labeling, the higher the credibility of the digital twin model.

[0045] Furthermore, the adaptive monitoring and scheduling submodule (215) selects to store complete raw data, key event fragments, feature values ​​or statistical summary data, so as to reduce the amount of invalid data storage while ensuring the integrity of key status information;

[0046] For the There are 1 monitoring channel, with a minimum sampling rate of 1. The highest sampling rate is Then the channel at time The sampling rate is expressed as:

[0047] ;

[0048] in, The sampling intensity factor has a value range of 0 to 1; the sampling intensity factor is expressed as:

[0049] ;

[0050] in, For the first The basic sampling weight of each monitoring channel under the current process semantic state. This is the adjustment coefficient for abnormal risks. This is a credibility adjustment coefficient for digital twin models. This is considered an abnormal risk. This represents the current technological semantic state. To assess the credibility of digital twin models.

[0051] Furthermore, the online self-verification submodule (216) is used to control the motion actuator of the manufacturing equipment to move to the location of the in-machine self-verification reference module (400) or the preset verification area of ​​the manufacturing object when the credibility of the digital twin model is lower than the preset threshold or the abnormal risk continues to be higher than the preset threshold, and to perform online reference measurement. The in-machine self-verification reference module (400) includes an in-machine reference part (410) and a reference part holder (420).

[0052] The in-machine reference component (410) is set in the non-operation area or safety verification area of ​​the manufacturing equipment and is fixed to the manufacturing equipment by the reference component holder (420). Its geometry is one or more of the following: standard sphere, reference plane, reference stepped groove, reference cylindrical surface, standard hole or standard edge. It is used to provide a repeatable and traceable reference measurement object in the online self-verification process.

[0053] Furthermore, in the online self-verification process of the online self-verification submodule (216), the reference value obtained from the reference measurement is set as follows: The system obtains the corresponding value based on sensor measurements or predictions from the equipment's digital twin model. Then refer to the residual Represented as:

[0054] ;

[0055] Subsequently, data analysis was performed on the residuals to determine their causes. When the reference residuals mainly manifested as sensor zero-point drift, the sensor zero-point parameters were corrected.

[0056] ;

[0057] in, and These are the sensor zero-point parameters before and after correction. This is a correction factor; when the reference residual mainly manifests as geometric mapping deviation, the geometric mapping parameters between the sensor coordinate system and the equipment coordinate system are corrected.

[0058] ;

[0059] in, and These are the geometric mapping parameters before and after the correction, respectively. The mapping correction amount is obtained based on the reference residual; when the reference residual mainly manifests as a deviation of the digital twin model, the edge parameters of the equipment digital twin model (212) are corrected:

[0060] ;

[0061] in, and The parameters of the edge-side equipment digital twin model before and after the correction are respectively. To equip the digital twin model with output functions, Adjust the gain matrix for the parameters;

[0062] When the reference residuals mainly manifest as a slow drift in the equipment baseline response, the mean vector and covariance matrix in the equipment baseline response model parameters are corrected based on the baseline response characteristics obtained from online reference measurements.

[0063] ;

[0064] ;

[0065] in, and These are the mean vectors of the equipment baseline response model before and after correction; and These are the covariance matrices before and after the correction, respectively. β represents the new baseline response characteristics obtained from online reference measurements; β is the model update coefficient.

[0066] Furthermore, the equipment baseline response normalization submodule (217) is used to establish the equipment baseline response model, assuming that the baseline response obtained during the standard calibration process is... Baseline response characteristics of the group, the first Baseline response characteristics of the group Then the mean vector and covariance matrix of the equipment baseline response model are expressed as follows:

[0067] ;

[0068] ;

[0069] Feature vectors acquired online Normalization is performed using the equipment baseline response model:

[0070] ;

[0071] in, This is the eigenvector after the baseline response is normalized.

[0072] A monitoring method for an intelligent monitoring system for the service status of manufacturing equipment is also provided, the monitoring method comprising the following steps:

[0073] S1: Complete the multi-source sensor deployment, equipment communication establishment, data sensing module (100) initialization, edge computing module (200) initialization, cloud module (300) communication establishment, and in-machine self-verification benchmark module (400) initialization;

[0074] S2: Perform equipment baseline response calibration by controlling the manufacturing equipment to run according to the preset no-load process parameter sequence, preset standard motion trajectory and preset reference part measurement sequence, to obtain equipment baseline response data and establish equipment baseline response model;

[0075] S3: Obtain process program, controller interpolation information, phase information of process execution components and position and attitude information of equipment motion execution mechanism, and combine with equipment digital twin model (212) to predict the equipment service conditions in the current and future preset time window, and generate process semantic state sequence and manufacturing object area number;

[0076] S4: Based on the process semantic state sequence, as well as the credibility and anomaly risk of the digital twin model in the previous monitoring cycle or initialization state, adaptive monitoring and scheduling are performed on the sampling rate, range, gain, channel enable status and buffer mode of the multi-source signal monitoring instrument (120), and the data storage granularity of event data.

[0077] S5: The data perception module (100) performs acquisition, conditioning, spatiotemporal unified marking and pre-trigger buffering of multi-source heterogeneous signals generated during the service of manufacturing equipment according to the scheduling results determined in step S4, and the process semantic parsing submodule (213) supplements the process semantic status number and manufacturing object area number to form an event data frame;

[0078] S6: The digital twin model credibility assessment submodule (214) calculates the credibility of the digital twin model for the current monitoring period based on the residual between the prediction results of the equipment digital twin model (212) and the measured values ​​of multi-source heterogeneous signals, sensor health status, spatiotemporal marker integrity and equipment baseline response offset.

[0079] S7: Input the event data frame processed by the equipment baseline response normalization submodule (217) into the intelligent reasoning model (211). The intelligent reasoning model (211) obtains the status of key equipment components, process status and corresponding manufacturing object area status reasoning results based on the process semantic status number and manufacturing object area number in the event data frame, and outputs the abnormal risk of the current monitoring cycle.

[0080] S8: When the credibility of the digital twin model in the current monitoring period is lower than the preset threshold or the abnormal risk continues to be higher than the preset threshold, the online self-verification submodule (216) determines the position of the on-machine self-verification benchmark module (400) or the preset verification area position of the manufacturing object according to the process semantic state number and manufacturing object area number corresponding to the abnormal event, and controls the motion actuator of the manufacturing equipment to move to the corresponding position to perform online reference measurement, and updates the sensor zero point, geometric mapping parameters, equipment baseline response model parameters and edge side equipment digital twin model parameters according to the measurement results;

[0081] S9: Upload the event data frame, digital twin model credibility sequence, abnormal risk sequence, reference measurement results, equipment baseline response data, edge-side intelligent reasoning model parameters and edge-side equipment digital twin model parameters to the cloud module (300), and classify and store them by the cloud storage submodule (310);

[0082] S10: When the cloud storage submodule (310) accumulates a preset number or preset period of event data, the cloud computing submodule (320) performs incremental learning update, transfer learning update or knowledge distillation update on the intelligent reasoning model (211) and equipment digital twin model (212) in the multi-source signal monitoring software (210) according to the event data, reference measurement results and edge-side intelligent reasoning model parameters and edge-side equipment digital twin model parameters, and sends the updated model parameters to the edge computing module (200) for state reasoning, digital twin prediction and adaptive monitoring scheduling in subsequent monitoring periods.

[0083] Furthermore, step S4 includes:

[0084] S41: When the current monitoring cycle is in standby mode, only the low sampling rate monitoring channel and the pre-trigger buffer unit (126) are turned on, and background data is collected at the first sampling rate;

[0085] S42: When the process semantic state enters the approach segment, or when the equipment digital twin model (212) predicts that the process execution component and the manufacturing object will come into contact, energy input, material removal or surface treatment within the future preset time window, switch to the pre-trigger mode, sample at a second sampling rate higher than the first sampling rate, and retain the cached data before triggering by the pre-trigger cache unit (126).

[0086] S43: When the process semantic state enters the process action establishment segment or steady-state operation segment, or when the abnormal risk exceeds the preset threshold, switch to full acquisition mode or abnormal enhanced acquisition mode, increase the sampling rate of at least one channel, and adjust the gain or range of the corresponding channel according to the channel signal amplitude, while enabling key event data storage.

[0087] S44: When the process semantic state enters the process action exit segment or rollback segment, the cached data after the trigger is retained, and when the abnormal risk returns to normal and the credibility of the digital twin model is higher than the preset threshold, the standby mode is returned. The credibility of the digital twin model and the abnormal risk after returning to the standby mode are used for the adaptive monitoring scheduling of the next monitoring cycle.

[0088] Step S8 includes:

[0089] S81: Based on the residual type of the digital twin model, the health status of the sensor, the semantic status of the process, and the trend of abnormal risk changes, make a preliminary judgment on whether the cause of self-verification is biased towards sensor drift, deviation of the digital twin model, or abnormality of the actual service status.

[0090] S82: If the judgment result is biased towards sensor drift or digital twin model deviation, the motion actuator of the control manufacturing equipment is moved to the position of the in-machine reference part (410) for reference measurement;

[0091] S83: If the judgment result is biased towards the abnormal actual service status, the online self-verification submodule (216) determines the preset verification area of ​​the manufacturing object in the preset safety verification window according to the manufacturing object area number corresponding to the abnormal event, and controls the motion actuator of the manufacturing equipment to move to the preset verification area of ​​the manufacturing object to perform local verification measurement.

[0092] S84: Update the sensor zero point, geometric mapping parameters, equipment baseline response model parameters, and edge-side equipment digital twin model parameters based on the reference measurement results, and recalculate the reliability and anomaly risk of the digital twin model;

[0093] S85: If the risk of anomalies still exceeds the preset threshold after the update, output an alarm, speed reduction, shutdown, or compensation suggestion.

[0094] The beneficial effects achieved by this invention are:

[0095] This invention uses a spatiotemporal unified labeling unit and a process semantic parsing submodule to associate multi-source heterogeneous signals with system time, phase information of process execution components, position and attitude information of equipment motion execution mechanisms, process semantic state, and manufacturing object area number, and encapsulates them into event data frames. This makes the collected data no longer just ordinary time series, but state data with information such as process stage, spatial location, sampling mode, model credibility, and model version, thereby improving the traceability of multi-source heterogeneous signals and the accuracy of subsequent state analysis.

[0096] This invention dynamically schedules the sampling rate, range, gain, channel enable status, caching mode, and data storage granularity of the data sensing module based on process semantic state, digital twin model credibility, and anomaly risk. Redundant acquisition is reduced during standby or low-risk phases. During the approach phase, process action establishment phase, steady-state operation phase, process action exit phase, or anomaly risk escalation phase, it enters pre-triggered, full-data acquisition, or anomaly-enhanced acquisition modes. A pre-triggered caching unit retains complete event data segments before, at, and after triggering, thereby improving the ability to capture transient information in critical states and reducing the acquisition and storage of low-value data.

[0097] This invention incorporates a self-verification benchmark module and combines digital twin model reliability assessment and equipment baseline response normalization methods. When monitoring results are uncertain or the risk of anomalies continues to rise, online reference measurements are performed in a preset verification area on the machine benchmark or manufacturing object to distinguish between sensor drift, digital twin model deviation, and abnormal actual service conditions. Furthermore, the sensor zero point, geometric mapping parameters, equipment baseline response model parameters, and equipment digital twin model parameters are corrected, thereby improving the reliability of monitoring results, the long-term adaptability of the model, and the system's cross-equipment application capability. Attached Figure Description

[0098] Appendix Figure 1 This is a schematic diagram of the overall structure of the intelligent monitoring system for the service status of manufacturing equipment.

[0099] Appendix Figure 2 This is a schematic diagram of signal distribution and fusion in a dual-path wide dynamic conditioning unit.

[0100] Appendix Figure 3 This is a schematic diagram of the spatiotemporal unified marker and event data frame structure.

[0101] Appendix Figure 4 This is a schematic diagram of the generation of event data segments in the pre-triggered cache unit.

[0102] Appendix Figure 5 This is a schematic diagram of process semantic state division and data association.

[0103] Appendix Figure 6 This is a schematic diagram of adaptive monitoring and scheduling state switching.

[0104] Appendix Figure 7 This is a flowchart of a method for intelligent monitoring of the service status of manufacturing equipment.

[0105] Appendix Figure 1-7 In the middle: 100—Data Sensing Module; 110—Multi-Source Sensor; 120—Multi-Source Signal Monitoring Instrument; 121—Data Acquisition Card; 122—Basic Signal Conditioning Module; 123—External Interface Unit; 124—Dual-Path Wide Dynamic Conditioning Unit; 125—Spatiotemporal Unified Marking Unit; 126—Pre-Trigger Buffer Unit; 200—Edge Computing Module; 210—Multi-Source Signal Monitoring Software; 211—Intelligent Inference Model; 212—Equipment Digital Twin Model; 213—Process Semantic Parsing Submodule; 214—Digital Twin Model Credibility Assessment Submodule; 215—Adaptive Monitoring and Scheduling Submodule; 216—Online Self-Verification Submodule; 217—Equipment Baseline Response Normalization Submodule; 300—Cloud Module; 310—Cloud Storage Submodule; 320—Cloud Computing Submodule; 400—On-Machine Self-Verification Baseline Module; 410—On-Machine Baseline Component; 420—Baseline Component Clamp. Detailed Implementation

[0106] The present invention will be further described below with reference to specific embodiments, and the advantages and features of the present invention will become clearer as a result. However, these embodiments are merely exemplary and do not constitute any limitation on the scope of the present invention. Those skilled in the art should understand that modifications or substitutions can be made to the details and form of the technical solutions of the present invention without departing from the spirit and scope of the present invention, but all such modifications and substitutions fall within the protection scope of the present invention.

[0107] As attached Figure 1As shown, the present invention provides an intelligent monitoring system for the service status of manufacturing equipment, including a data sensing module 100, an edge computing module 200, a cloud module 300, and an on-machine self-verification benchmark module 400. The data sensing module 100 is used to acquire multi-source heterogeneous signals generated during the service of manufacturing equipment; the edge computing module 200 is used to receive and process the multi-source heterogeneous signals acquired by the data sensing module 100, generate process semantic status, digital twin model credibility, and anomaly risk, and schedule the sampling process, conditioning process, caching process, event data storage process, and on-machine self-verification process of the data sensing module 100 according to the process semantic status, digital twin model credibility, and anomaly risk; the cloud module 300 is used to store manufacturing equipment service status event data, digital twin model credibility sequence, anomaly risk sequence, reference measurement results, equipment baseline response data, edge-side intelligent inference model parameters, edge-side equipment digital twin model parameters, and edge-side model version information, and update the edge-side model parameters; the on-machine self-verification benchmark module 400 is used to provide reference geometric dimensions, reference spatial positions, or reference measurement values ​​when the edge computing module 200 issues a self-verification command, so as to perform online verification of sensor status, equipment digital twin model status, and the credibility of monitoring results.

[0108] The data sensing module 100 includes a multi-source sensor 110 and a multi-source signal monitoring instrument 120. The multi-source sensor 110 is used to acquire multi-source heterogeneous signals, including two or more signals from temperature, vibration, force, acoustic emission, displacement, rotational speed, current, and power. The multi-source signal monitoring instrument 120 is an integrated hardware enclosure, including a data acquisition card 121, a basic signal conditioning module 122, an external interface unit 123, a dual-path wide dynamic range conditioning unit 124, a spatiotemporal unified marking unit 125, and a pre-trigger buffer unit 126. The basic signal conditioning module 122 is used to perform one or more processing operations on the output signals of different types of sensors, including sensor excitation, impedance matching, filtering, isolation protection, charge conversion, or voltage conversion. The external interface unit 123 provides one or more of the following interfaces: a multi-source sensor interface, a communication interface, a power interface, a trigger synchronization interface, and a grounding shield interface, to achieve a unified connection between the multi-source sensor 110 and the multi-source signal monitoring instrument 120. Therefore, the multi-source signal monitoring instrument 120 can uniformly access signals of different types, amplitudes, and sampling frequencies, and provide a hardware foundation for subsequent dynamic scheduling and event data frame generation.

[0109] The edge computing module 200 runs multi-source signal monitoring software 210. The multi-source signal monitoring software 210 includes an intelligent inference model 211, an equipment digital twin model 212, a process semantic parsing submodule 213, a digital twin model credibility assessment submodule 214, an adaptive monitoring scheduling submodule 215, an online self-verification submodule 216, and an equipment baseline response normalization submodule 217. The intelligent inference model 211 is used to generate inference results on the status of key equipment components, process status, or manufacturing object status based on the processed multi-source heterogeneous signals through algorithmic inference, and outputs anomaly risks. The equipment digital twin model 212 is used to predict the service conditions of the manufacturing equipment within the current and future preset time windows based on the process program, controller interpolation information, phase information of process execution components, position and attitude information of equipment motion execution mechanisms, and geometric information of the manufactured object, and generate predicted service conditions. Among them, the process program and controller interpolation information are obtained by the CNC system of the manufacturing equipment; the phase information of process execution components is fed back by the encoder or controller of the manufacturing equipment; the position and attitude information of equipment motion execution mechanisms is fed back by the CNC system or controller of the manufacturing equipment; and the geometric information of the manufactured object is obtained by the process model or machining program.

[0110] Predicted service conditions include the motion state, process operation state, spatial position state, phase state, and load or energy input state of the manufacturing equipment during service. Among them, the motion state includes spindle speed, process execution component speed, feed rate, or relative motion speed; the process operation state includes the proximity distance, contact state, and action margin between the process execution component and the manufactured object, as well as contact, energy input, material removal, surface treatment, process operation establishment, steady-state operation, and process operation exit; the spatial position state includes the position and attitude information of the equipment motion execution mechanism, process head pose, trajectory point information, and manufactured object area number; the phase state includes the spindle phase, turntable phase, or other process execution component phase information; and the load or energy input state includes force, load, current, power, acoustic emission, vibration, or temperature response characteristics.

[0111] As attached Figure 2 As shown, the dual-path wide dynamic range conditioning unit 124 is disposed between the multi-source sensor 110 and the data acquisition card 121, and includes a signal distribution circuit, a high-gain conditioning link, and a low-gain conditioning link. For the sensor-acquired signals... The signal distribution circuit allocates it to the high-gain conditioning link and the low-gain conditioning link. The output signal of the high-gain conditioning link can be expressed as:

[0112] (1)

[0113] The output signal of a low-gain conditioning link can be expressed as:

[0114] (2)

[0115] in, This refers to the amplification factor of a high-gain link. This is the amplification factor for a low-gain link, and ; and These represent the noise in the two conditioning links. The high-gain link is used to enhance weak state characteristics, while the low-gain link is used to preserve strong transient signals. When the high-gain link is not saturated, the equivalent signal after weak characteristic enhancement can be obtained according to the following formula:

[0116] (3)

[0117] When a high-gain link becomes saturated, the equivalent signal with strong transient fidelity can be obtained according to the following formula:

[0118] (4)

[0119] If both signals are within their effective operating range, the edge computing module 200 can fuse the two signals based on noise levels and saturation conditions. The fused signal is represented as follows:

[0120] (5)

[0121] in, and The weights for the high-gain and low-gain links are respectively determined by the channel noise variance, saturation state, and anomaly risk. In this way, the dual-path wide dynamic range conditioning unit 124 can simultaneously enhance weak state characteristics and preserve strong transient signals during the same monitoring process, thereby obtaining wide dynamic range monitoring data.

[0122] As attached Figure 3As shown, the spatiotemporal unified marking unit 125 includes a clock reference circuit, a counter circuit, a phase information acquisition interface for process execution components, a position and attitude information acquisition interface for equipment motion execution mechanisms, and a trigger synchronization interface. The clock reference circuit provides a unified time reference to the data acquisition card 121 and the edge computing module 200; the counter circuit acquires pulses from the spindle encoder, turntable encoder, speed sensor, or process cycle trigger signals, and calculates the phase information of the process execution components; the position and attitude information acquisition interface for the equipment motion execution mechanisms acquires equipment axis coordinates, process head pose, feed status, trajectory point information, or controller interpolation information; the trigger synchronization interface sends a unified trigger signal to different acquisition channels, enabling multi-source heterogeneous signals to be stored corresponding to system time, process execution component phase information, and equipment motion execution mechanism position and attitude information. Simultaneously, the spatiotemporal unified marking unit 125 calculates the spatiotemporal marking integrity based on the number of channels that have completed valid spatiotemporal marking in the current event data frame and the total number of channels that should have completed marking.

[0123] Let the first The system time at each sampling moment is The phase information of the process execution component is The position and attitude information of the equipment's motion actuators are The process semantic status number is The manufacturing object area number is The sampling mode number is The credibility of the digital twin model is The abnormal risk is The model version number is Multi-source data vectors are Then the edge computing module 200 can encapsulate the spatiotemporally tagged data into event data frames:

[0124] (6)

[0125] in, It can be composed of one or more signals from temperature, vibration, force, acoustic emission, displacement, rotational speed, current, and power. Therefore, the event data frame not only includes sensor acquisition values, but also the corresponding system time, phase information of process execution components, equipment spatial location, process semantic state, manufacturing object area, sampling mode, model confidence level, and model version information. This enables multi-source heterogeneous signals to be transformed from ordinary time-series data into event-based data with process semantics and model state identifiers.

[0126] in, The manufacturing object region number corresponding to the k-th sampling time is used to represent the multi-source data vector. The corresponding local area of ​​the manufactured object. By... By writing event data frames, the edge computing module 200 and the cloud module 300 can retrieve, classify and store, infer state, trace anomalies, and determine online self-verification locations of event data according to the manufacturing object region.

[0127] As attached Figure 4 As shown, the pre-trigger buffer unit 126 includes a ring buffer and a trigger latch circuit. The ring buffer is used to continuously buffer multi-source signal historical data for a preset time length in standby mode or pre-trigger mode. Let the trigger time be... The cache time before triggering is The length of time to save after triggering is Then the complete event data segment generated by the pre-trigger cache unit 126 can be represented as:

[0128] (7)

[0129] in, For multi-source signal data, This corresponds to the event data frame. The trigger latch circuit is used to lock the cached data before the trigger moment and continue recording the subsequent edge data after the trigger moment when the process semantic state enters the approach segment, process action establishment segment, or process action exit segment, or when the abnormal risk exceeds a preset threshold, or when the credibility of the digital twin model is lower than a preset threshold. Accordingly, the trigger condition can be expressed as:

[0130] (8)

[0131] or

[0132] (9)

[0133] or

[0134] (10)

[0135] in, , , These represent the approach stage, the process action establishment stage, and the process action exit stage, respectively. This is the threshold for abnormal risk. This sets the credibility threshold for the digital twin model. Through this method, the system can save data before triggering, data at the triggering moment, and data after triggering, preventing the omission of effective information before and after key state changes.

[0136] As attached Figure 5As shown, the process semantic parsing submodule 213 is used to divide the service process of manufacturing equipment into two or more of the following: standby phase, approach phase, process action establishment phase, steady-state operation phase, process action exit phase, rollback phase, and online detection phase. Let the feature vector used for process semantic judgment be:

[0137] (11)

[0138] in, To equip the digital twin model 212 with the predicted proximity distance, contact state, or operational margin between the process execution components and the manufactured object, This refers to the feed rate or relative motion speed. Spindle speed or speed of the process execution component. For the phase information of the process execution components, To provide position and attitude information for the motion actuators, Characteristics of force signals or load signals. These are characteristics of acoustic emission, vibration, current, power, or other process states. For the set of process semantic states:

[0139] (12)

[0140] The probability of the current time belonging to the i-th time step can be calculated using the following probabilistic form. The probability of each process semantic state:

[0141] (13)

[0142] in, This is the process semantic state number for which the probability is currently to be calculated. and The first The weight vector and bias corresponding to each process semantic state; This represents the total number of process semantic states. For summing indexes of process semantic states, and The first The weight vector and bias corresponding to each process semantic state. The process semantic state at the current moment can be determined as:

[0143] (14)

[0144] Therefore, the system can map the service process of manufacturing equipment to a specific process semantic state based on the predicted service conditions, the position and attitude information of the equipment motion actuator, the phase information of the process execution components, and multi-source heterogeneous signals. Furthermore, it can generate a manufacturing object area number to characterize the local processing area, surface treatment area, detection area, or preset verification area of ​​the manufacturing object corresponding to the current monitoring data. This allows the multi-source heterogeneous signals to correspond not only to the process time and equipment motion state, but also to the specific spatial area of ​​the manufacturing object.

[0145] The digital twin model credibility assessment submodule 214 is used to calculate the credibility of the digital twin model based on the residuals between the prediction results of the equipment digital twin model 212 and the measured results of multi-source heterogeneous signals, sensor health status, spatiotemporal marker integrity, and equipment baseline response offset. Let the equipment digital twin model 212 at time... The output predicted feature vector is The measured feature vector, collected and processed by the data sensing module 100, is: The weighted residual between the two can be expressed as:

[0146] (15)

[0147] in, Let be the residual weight matrix, used to adjust the contribution of different signal features to the residual. Let the be... The health status of each sensor is Its weight is The health status of the sensor can then be expressed as:

[0148] (16)

[0149] in, Let be the number of sensors participating in the monitoring. Assume the number of channels in the current event data frame that have completed valid spatiotemporal marking is . The total number of channels that should complete spatiotemporal marking is Then, the spatiotemporal label integrity can be expressed as:

[0150] (17)

[0151] Furthermore, the equipment baseline response offset is used to characterize the degree of deviation of the current equipment baseline response relative to the pre-established equipment baseline response model. Let the current baseline response characteristics be... The mean vector of the equipment baseline response model is The covariance matrix is Then the equipment baseline response offset can be expressed as:

[0152] (18)

[0153] The larger the prediction residual and the larger the equipment baseline response offset, the lower the reliability of the digital twin model; the better the sensor health status and the more complete the spatiotemporal labeling, the higher the reliability of the digital twin model. Therefore, the reliability of the digital twin model can be defined as:

[0154] (19)

[0155] in, and The formula indicates that the reliability of the digital twin model is jointly determined by the model prediction residuals, sensor health status, spatiotemporal label integrity, and equipment baseline response offset, and can reflect the reliability of the equipment digital twin model 212 in representing the actual equipment status under the current operating conditions.

[0156] The intelligent reasoning model 211 is used to obtain reasoning results on the status of key equipment components, process status, or manufacturing object status based on processed multi-source heterogeneous signals, and outputs the anomaly risk. Let the anomaly probability output by the intelligent reasoning model 211 be... The abrupt change in multi-source signals is The risk weight corresponding to the current process semantic state is: Then, the abnormal risk can be expressed as:

[0157] (20)

[0158] in, For the Sigmoid function, , where is the weighting coefficient. This formula indicates that when the anomaly probability output by the intelligent inference model 211 increases, the credibility of the digital twin model decreases, the abrupt change of multi-source signals increases, or the current process semantic state is in a stage prone to transient changes, such as the approach stage, the process action establishment stage, or the process action exit stage, the anomaly risk increases accordingly.

[0159] The adaptive monitoring and scheduling submodule 215 is used to dynamically schedule the sampling, conditioning, and caching status of the data perception module 100, as well as the storage granularity of event data, based on the process semantic status, the reliability of the digital twin model, and the risk of anomalies. For the first... There are 1 monitoring channel, with a minimum sampling rate of 1. The highest sampling rate is Then the channel at time The sampling rate can be expressed as:

[0160] (twenty one)

[0161] in, The sampling intensity factor ranges from 0 to 1. The sampling intensity factor can be expressed as:

[0162] (twenty two)

[0163] in, For the first The basic sampling weight of each monitoring channel under the current process semantic state. This is the adjustment coefficient for abnormal risks. This is the credibility adjustment coefficient for the digital twin model. Therefore, when the manufacturing equipment is in critical process semantic states such as the approach stage, process action establishment stage, or process action exit stage, or when the anomaly risk is high and the credibility of the digital twin model is low, the system automatically increases the sampling intensity of relevant channels; when the manufacturing equipment is in standby or low-risk stages, the system reduces the sampling rate and the storage granularity of event data, thereby reducing the collection and storage of low-value data.

[0164] The storage granularity of the event data may include one or more storage levels among raw waveform data, downsampled data, eigenvalue data, statistical summary data, complete event data segments, and model input samples. The adaptive monitoring and scheduling submodule 215, based on the process semantic state, the reliability of the digital twin model, and the anomaly risk, first determines the sampling rate, range, gain, channel enable status, buffering mode, and event data storage granularity of each monitoring channel, and then sends the corresponding scheduling parameters to the data sensing module 100 to control the corresponding monitoring channel to perform data acquisition, signal conditioning, buffering, and storage according to the scheduling results.

[0165] When the system is in standby mode, a low sampling rate and low storage granularity strategy can be adopted. When the process semantic state enters the approach stage, process action establishment stage, steady-state operation stage, or abnormal risk increases, the sampling rate of relevant monitoring channels can be increased, the range and gain can be adjusted, the corresponding channels can be opened, and the pre-trigger buffer or full acquisition mode can be switched. At the same time, complete raw data, key event fragments, feature values, or statistical summary data can be stored according to monitoring needs. When the system recovers to a low-risk state, a lower sampling rate and lower storage granularity are restored. Thus, while ensuring the integrity of key status information, the amount of invalid data collection and storage is reduced, as shown in the appendix. Figure 6As shown, in standby mode, only the low sampling rate monitoring channel and the pre-trigger cache unit 126 are activated, and background data is collected at the first sampling rate. When the process semantic state enters the approach phase, or when the equipment digital twin model 212 predicts that the process execution component and the manufacturing object will come into contact, receive energy, remove materials, or undergo surface treatment within a preset time window, the system switches to the pre-trigger mode and samples at a second sampling rate higher than the first sampling rate. The pre-trigger cache unit 126 retains the cached data before the trigger. When the process semantic state enters the process action establishment phase or steady-state operation phase, or when the abnormal risk exceeds a preset threshold, the system switches to the full acquisition mode or the abnormal enhanced acquisition mode, increases the sampling rate of at least one channel, and adjusts the gain or range of the corresponding channel according to the channel signal amplitude. At the same time, the system activates the key event data storage. When the process semantic state enters the process action exit phase or rollback phase, the cached data after the trigger is retained. When the abnormal risk returns to normal and the credibility of the digital twin model is higher than the preset threshold, the system returns to standby mode. The credibility of the digital twin model and the abnormal risk after returning to standby mode are used for adaptive monitoring scheduling in the next monitoring cycle.

[0166] The equipment baseline response normalization submodule 217 is used to establish the equipment baseline response model and perform baseline response normalization processing on the multi-source heterogeneous signals acquired online. Specifically, based on the reference responses acquired by the manufacturing equipment under preset no-load process parameter sequences, preset standard motion trajectories, and preset reference component measurement sequences, the system establishes one or more of the following baselines for the manufacturing equipment: temperature rise baseline, vibration transmission baseline, phase lag baseline of process execution components, geometric error baseline, load response baseline, and energy input response baseline. (The data is obtained during the standard calibration process.) Baseline response characteristics of the group, the first Baseline response characteristics of the group Then the mean vector and covariance matrix of the equipment baseline response model can be expressed as follows:

[0167] (twenty three)

[0168] (twenty four)

[0169] Feature vectors acquired online Normalization can be performed using the equipment baseline response model:

[0170] (25)

[0171] in, This is the eigenvector after the baseline response is normalized.

[0172] The equipment baseline response offset is calculated by the difference between the equipment baseline response model and the current baseline response characteristics. Through the above processing, the data distribution offset caused by differences in temperature rise baseline, vibration transmission characteristics, phase lag of process execution components, geometric error baseline, load response baseline and energy input response baseline between different manufacturing equipment can be reduced, so that the intelligent inference model 211 and the equipment digital twin model 212 have better cross-equipment adaptability.

[0173] The online self-verification submodule 216 is used to control the motion actuator of the manufacturing equipment to move to the location of the on-machine self-verification reference module 400 or the preset verification area of ​​the manufactured object when the credibility of the digital twin model is lower than a preset threshold or the abnormal risk continues to be higher than a preset threshold, and to perform online reference measurement. The on-machine self-verification reference module 400 includes an on-machine reference component 410 and a reference component holder 420. The on-machine reference component 410 is set in the non-operation area or safety verification area of ​​the manufacturing equipment and is fixed to the manufacturing equipment by the reference component holder 420. Its geometry is one or more of a standard sphere, a reference plane, a reference stepped groove, a reference cylindrical surface, a standard hole, or a standard edge, and is used to provide a repeatable and traceable reference measurement object during the online self-verification process.

[0174] During the online self-verification process, the reference value obtained from the reference measurement is set as follows: The system obtains the corresponding value based on sensor measurements or predictions from the equipment's digital twin model. Then the reference residual can be expressed as:

[0175] (26)

[0176] Subsequently, data analysis can be performed on the residuals to determine their causes. When the reference residuals mainly manifest as sensor zero-point drift, the sensor zero-point parameters can be corrected.

[0177] (27)

[0178] in, and These are the sensor zero-point parameters before and after correction. This is a correction factor. When the reference residual mainly manifests as a geometric mapping deviation, the geometric mapping parameters between the sensor coordinate system and the equipment coordinate system can be corrected.

[0179] (28)

[0180] in, and These are the geometric mapping parameters before and after the correction, respectively. This is the mapping correction amount obtained from the reference residual. When the reference residual mainly manifests as a slow drift in the equipment baseline response, the mean vector and covariance matrix in the equipment baseline response model parameters can be corrected based on the baseline response characteristics obtained from online reference measurements.

[0181] (29)

[0182] (30)

[0183] in, and These are the mean vectors of the equipment baseline response model before and after correction; and These are the covariance matrices before and after the correction, respectively. The new baseline response characteristics are obtained from online reference measurements; β is the model update coefficient. The corrected equipment baseline response model parameters can be used for subsequent equipment baseline response normalization processing and cross-equipment applications. When the reference residuals mainly manifest as deviations from the digital twin model, the edge parameters of the equipment digital twin model 212 are corrected:

[0184] (31)

[0185] in, and The parameters of the edge-side equipment digital twin model before and after the correction are respectively. To equip the digital twin model with output functions, The gain matrix is ​​adjusted for the parameters. Thus, the system distinguishes between sensor drift, digital twin model deviation, and abnormal actual service conditions based on online reference measurement results, and outputs alarms, speed reduction, shutdown, or compensation suggestions when necessary.

[0186] The cloud module 300 includes a cloud storage submodule 310 and a cloud computing submodule 320.

[0187] The cloud storage submodule 310 is used to store monitoring parameters, including manufacturing equipment service status event data, digital twin model credibility sequence, abnormal risk sequence, reference measurement results, equipment baseline response data, edge-side intelligent inference model parameters, edge-side equipment digital twin model parameters, and edge-side model version information.

[0188] Among them, the service status event data of manufacturing equipment is formed by the collaboration of multiple modules. The data sensing module 100 collects multi-source heterogeneous signals; then the spatiotemporal unified marking unit 125 completes the unified time marking, the association of phase information of process execution components and the association of position and attitude information of equipment motion execution mechanism; then the process semantic parsing submodule 213 supplements the process semantic status and manufacturing object area number; finally, it is encapsulated to form an event data frame.

[0189] The credibility sequence of the digital twin model is a credibility sequence formed by the credibility of the digital twin model for each monitoring period.

[0190] The abnormal risk sequence is formed by the abnormal risk values ​​of each monitoring period.

[0191] The reference measurement results are obtained by the online self-verification submodule 216 controlling the manufacturing equipment to perform online reference measurements.

[0192] The equipment baseline response data is used to establish the equipment baseline response model by the equipment baseline response normalization submodule; during subsequent monitoring, the current equipment baseline response characteristics are continuously recorded and used for equipment baseline response model updates and cross-equipment normalization processing.

[0193] The parameters of the edge-side intelligent inference model are built into the intelligent inference model 211. The cloud can read the current parameters of the edge model and perform incremental learning, transfer learning, or knowledge distillation updates.

[0194] The parameters of the edge-side equipment digital twin model are built into the equipment digital twin model 212. The cloud uses the uploaded data to update the digital twin model and then redistributes the updated model parameters to the edge side.

[0195] The edge-side model version information is recorded after each edge-side model parameter update or cloud-based model parameter distribution. The version information corresponding to the current intelligent inference model and equipment digital twin model is recorded and uploaded to the cloud for model version management and historical traceability.

[0196] The cloud computing submodule 320 is used to perform incremental learning updates, transfer learning updates, or knowledge distillation updates on the intelligent inference model 211 and the equipment digital twin model 212 in the multi-source signal monitoring software 210 based on the event data, reference measurement results, edge-side intelligent inference model parameters, and edge-side equipment digital twin model parameters in the cloud storage submodule 310, and then send the updated model parameters to the edge computing module 200.

[0197] Let the first The edge-side model parameters uploaded by the Taiwanese manufacturing equipment are The corresponding number of event data is The number of manufacturing equipment participating in cloud updates is Then, the cloud can use a weighted aggregation method to update the global model parameters:

[0198] (32)

[0199] in, These are the updated model parameters from the cloud. Furthermore, the cloud computing submodule 320 can also perform model updates using the following objective function:

[0200] (33)

[0201] in, For state recognition or prediction task loss, For the consistency constraint loss of digital twins, To update the model parameters, and These are the weighting coefficients. In this way, the cloud module 300 can utilize newly added event data to evolve the model while preserving the stability of the original model, and then send the updated model parameters to the edge computing module 200.

[0202] As attached Figure 7 As shown, the present invention also provides an intelligent monitoring method for the service status of manufacturing equipment, comprising the following steps:

[0203] Step S1: Complete the multi-source sensor deployment, equipment communication establishment, data sensing module 100 initialization, edge computing module 200 initialization, cloud module 300 communication establishment, and on-board self-verification benchmark module 400 initialization;

[0204] Step S2: Perform equipment baseline response calibration. By controlling the manufacturing equipment to run according to the preset no-load process parameter sequence, preset standard motion trajectory and preset reference part measurement sequence, obtain equipment baseline response data and establish equipment baseline response model;

[0205] Step S3: Obtain the process program, controller interpolation information, phase information of process execution components, and position and attitude information of equipment motion execution mechanism. Combine the equipment digital twin model 212 to predict the equipment service conditions within the current and future preset time windows, and generate process semantic state sequence and manufacturing object area number.

[0206] Step S4: Based on the process semantic state sequence, as well as the credibility and anomaly risk of the digital twin model in the previous monitoring cycle or initialization state, adaptive monitoring and scheduling are performed on the sampling rate, range, gain, channel enable status and buffer mode of the multi-source signal monitoring instrument 120, and the data storage granularity of event data.

[0207] Step S4 includes the following steps:

[0208] S41: When the current monitoring cycle is in standby mode, only the low sampling rate monitoring channel and the pre-trigger buffer unit are turned on, and background data is collected at the first sampling rate.

[0209] S42: When the process semantic state enters the approach phase, or when the equipment digital twin model predicts that the process execution component and the manufacturing object will come into contact, receive energy, remove materials or perform surface treatment within a future preset time window, switch to the pre-trigger mode, sample at a second sampling rate higher than the first sampling rate, and retain the cached data before triggering by the pre-trigger cache unit.

[0210] S43: When the process semantic state enters the process action establishment segment or steady-state operation segment, or when the abnormal risk exceeds the preset threshold, switch to full acquisition mode or abnormal enhanced acquisition mode, increase the sampling rate of at least one channel, and adjust the gain or range of the corresponding channel according to the channel signal amplitude, while enabling key event data storage.

[0211] S44: When the process semantic state enters the process action exit segment or rollback segment, the cached data after the trigger is retained, and when the abnormal risk returns to normal and the credibility of the digital twin model is higher than the preset threshold, the standby mode is returned. The credibility of the digital twin model and the abnormal risk after returning to the standby mode are used for the adaptive monitoring scheduling of the next monitoring cycle.

[0212] Step S5: The data perception module 100 performs acquisition, conditioning, spatiotemporal unified marking and pre-trigger buffering of multi-source heterogeneous signals generated during the service of manufacturing equipment according to the scheduling result determined in step S4, and the process semantic parsing submodule 213 supplements the process semantic status number and manufacturing object area number to form an event data frame.

[0213] Step S6: The digital twin model credibility assessment submodule 214 calculates the credibility of the digital twin model for the current monitoring period based on the residual between the prediction results of the equipment digital twin model 212 and the measured values ​​of multi-source heterogeneous signals, sensor health status, spatiotemporal marker integrity, and equipment baseline response offset.

[0214] Step S7: Input the event data frame processed by the equipment baseline response normalization submodule 217 into the intelligent reasoning model 211. The intelligent reasoning model 211 obtains the status of key equipment components, process status and corresponding manufacturing object area status reasoning results based on the process semantic status number and manufacturing object area number in the event data frame, and outputs the abnormal risk of the current monitoring period.

[0215] Step S8: When the credibility of the digital twin model in the current monitoring period is lower than the preset threshold or the abnormal risk continues to be higher than the preset threshold, the online self-verification submodule 216 determines the position of the on-machine self-verification benchmark module 400 or the preset verification area position of the manufacturing object based on the process semantic state number and manufacturing object area number corresponding to the abnormal event, and controls the motion actuator of the manufacturing equipment to move to the corresponding position to perform online reference measurement. Based on the measurement results, the sensor zero point, geometric mapping parameters, equipment baseline response model parameters and edge-side equipment digital twin model parameters are updated.

[0216] Step S8 includes the following steps:

[0217] S81: Based on the residual type of the digital twin model, the health status of the sensor, the semantic status of the process, and the trend of abnormal risk changes, make a preliminary judgment on whether the cause of self-verification is biased towards sensor drift, deviation of the digital twin model, or abnormality of the actual service status.

[0218] S82: If the judgment result is biased towards sensor drift or digital twin model deviation, control the motion actuator of the manufacturing equipment to move to the position of the in-machine reference part for reference measurement;

[0219] S83: If the judgment result is biased towards the abnormal actual service status, the online self-verification submodule 216 determines the preset verification area of ​​the manufacturing object in the preset safety verification window according to the manufacturing object area number corresponding to the abnormal event, and controls the motion actuator of the manufacturing equipment to move to the preset verification area of ​​the manufacturing object to perform local verification measurement.

[0220] S84: Update the sensor zero point, geometric mapping parameters, equipment baseline response model parameters, and edge-side equipment digital twin model parameters based on the reference measurement results, and recalculate the reliability and anomaly risk of the digital twin model;

[0221] S85: If the risk of anomalies still exceeds the preset threshold after the update, output an alarm, speed reduction, shutdown, or compensation suggestion.

[0222] Step S9: Upload the event data frame, digital twin model credibility sequence, abnormal risk sequence, reference measurement results, equipment baseline response data, edge-side intelligent inference model parameters, and edge-side equipment digital twin model parameters to the cloud module 300, and classify and store them by the cloud storage submodule 310.

[0223] Step S10: After the cloud storage submodule 310 accumulates a preset number or preset period of event data, the cloud computing submodule 320 performs incremental learning update, transfer learning update or knowledge distillation update on the intelligent inference model 211 and the equipment digital twin model 212 in the multi-source signal monitoring software 210 according to the event data, reference measurement results, edge-side intelligent inference model parameters and edge-side equipment digital twin model parameters, and sends the updated model parameters to the edge computing module 200 for state inference, digital twin prediction and adaptive monitoring scheduling in subsequent monitoring periods.

[0224] The above are merely preferred embodiments of the present invention and do not constitute any limitation on the scope of protection of the present invention; all technical solutions formed by equivalent transformations or equivalent substitutions fall within the scope of protection of the present invention; the parts of the present invention not described in detail are well known to those skilled in the art.

Claims

1. An intelligent monitoring system for the service status of manufacturing equipment, characterized in that, The intelligent monitoring system for the service status of manufacturing equipment includes: The data sensing module (100) is used to acquire multi-source heterogeneous signals generated during the service of manufacturing equipment; the multi-source heterogeneous signals include two or more signals from temperature, vibration, force, acoustic emission, displacement, rotational speed, current and power; The edge computing module (200) runs multi-source signal monitoring software (210), which includes: The equipment digital twin model (212) is used to generate predicted service conditions for the current and future preset time windows of the manufacturing equipment; The process semantic parsing submodule (213) is used to divide the service process of manufacturing equipment into different process semantic states based on the predicted service conditions, the position and attitude information of the equipment motion actuator, the phase information of the process execution component and the multi-source heterogeneous signal; the process semantic states include standby segment, approach segment, process action establishment segment, steady-state operation segment, process action exit segment, retreat segment and online detection segment; The digital twin model credibility assessment submodule (214) is used to calculate the credibility of the digital twin model based on the predicted service conditions; The intelligent reasoning model (211) is used to output anomaly risks based on the credibility of the digital twin model; The adaptive monitoring and scheduling submodule (215) is used to dynamically schedule the storage granularity based on the process semantic state, the credibility of the digital twin model, and the risk of anomalies. The adaptive monitoring and scheduling submodule (215) selects to store complete raw data, key event fragments, feature values ​​or statistical summary data to reduce the amount of invalid data storage while ensuring the integrity of key status information; For the There are 1 monitoring channel, with a minimum sampling rate of 1. The highest sampling rate Then the channel at time The sampling rate is expressed as: ； in, The sampling intensity factor has a value range of 0 to 1; the sampling intensity factor is expressed as: ； in, For the first The basic sampling weight of each monitoring channel under the current process semantic state. This is the adjustment coefficient for abnormal risks. This is a credibility adjustment coefficient for digital twin models. This is considered an abnormal risk. This represents the current technological semantic state. To assess the credibility of digital twin models; The online self-verification submodule (216) is used to perform online reference measurements when the credibility of the digital twin model is lower than a preset threshold or the abnormal risk is consistently higher than a preset threshold, and to correct the sensor and model parameters based on the measurement results; The equipment baseline response normalization submodule (217) is used to establish the equipment baseline response model and perform baseline response normalization processing on the multi-source heterogeneous signals acquired online. The cloud module (300) is used to store monitoring parameters and update the edge-side model parameters, and then send the updated model parameters to the edge computing module (200). The on-machine self-verification benchmark module (400) is used to perform online verification when the edge computing module (200) issues a self-verification command.

2. The intelligent monitoring system for the service status of manufacturing equipment as described in claim 1, characterized in that, The abnormal risks Represented as: ； in, This represents the probability of an anomaly. For the abrupt change of multi-source signals, The risk weight corresponding to the current process semantic state. To enhance the credibility of digital twin models, For the Sigmoid function, These are the weighting coefficients.

3. The intelligent monitoring system for the service status of manufacturing equipment as described in claim 1, characterized in that, The feature vector used for process semantic judgment is: ； in, To equip the digital twin model (212) with the predicted proximity distance, contact state, or operational margin between the process execution component and the manufactured object, This refers to the feed rate or relative motion speed. Spindle speed or speed of the process execution component. For the phase information of the process execution components, To provide position and attitude information for the motion actuators, Characteristics of force signals or load signals. Characteristics of acoustic emission, vibration, current, or power; For the set of process semantic states: ； The probability of the current time belonging to the i-th time step is calculated using the following probabilistic form. The probability of each process semantic state: ； in, This is the process semantic state number for which the probability is currently to be calculated. and The first The weight vector and bias corresponding to each process semantic state; This represents the total number of process semantic states. For summing indexes of process semantic states, and The first The weight vector and bias corresponding to each process semantic state; The current process semantic state is: ； This represents the current technological semantic state.

4. The intelligent monitoring system for the service status of manufacturing equipment as described in claim 1, characterized in that, Credibility of digital twin models Defined as: ； in, and This is the adjustment coefficient; The residual between predicted service conditions and measured results of multi-source heterogeneous signals Represented as: ； in, To equip the digital twin model (212) at time The output predicted feature vector, The measured feature vectors of multi-source heterogeneous signals are... This is the residual weight matrix; Let the first The health status of each sensor is Its weight is Then the sensor health status Represented as: ； in, Let be the number of sensors participating in the monitoring; let the number of channels in the current event data frame that have completed valid spatiotemporal marking be . The total number of channels that should complete spatiotemporal marking is Then the spatiotemporal marking integrity Represented as: ； The equipment baseline response offset is used to characterize the degree of deviation of the current equipment baseline response relative to the pre-established equipment baseline response model; let the current baseline response characteristics be... The mean vector of the equipment baseline response model is The covariance matrix is Then the equipment baseline response offset Represented as: ； The larger the prediction residual and the larger the equipment baseline response offset, the lower the credibility of the digital twin model; the better the sensor health status and the more complete the spatiotemporal labeling, the higher the credibility of the digital twin model.

5. The intelligent monitoring system for the service status of manufacturing equipment as described in claim 1, characterized in that, The online self-verification submodule (216) is used to control the motion actuator of the manufacturing equipment to move to the location of the in-machine self-verification reference module (400) or the preset verification area of ​​the manufacturing object when the credibility of the digital twin model is lower than the preset threshold or the abnormal risk continues to be higher than the preset threshold, and to perform online reference measurement. The in-machine self-verification reference module (400) includes an in-machine reference part (410) and a reference part holder (420). The in-machine reference component (410) is set in the non-operation area or safety verification area of ​​the manufacturing equipment and is fixed to the manufacturing equipment by the reference component holder (420). Its geometry is one or more of the following: standard sphere, reference plane, reference stepped groove, reference cylindrical surface, standard hole or standard edge. It is used to provide a repeatable and traceable reference measurement object in the online self-verification process.

6. The intelligent monitoring system for the service status of manufacturing equipment as described in claim 1, characterized in that, In the online self-verification submodule (216), during the online self-verification process, the reference value obtained from the reference measurement is set to... The system obtains the corresponding value based on sensor measurements or predictions from the equipment's digital twin model. Then refer to the residual Represented as: ； Subsequently, data analysis was performed on the residuals to determine their causes. When the reference residuals mainly manifested as sensor zero-point drift, the sensor zero-point parameters were corrected. ； in, and These are the sensor zero-point parameters before and after correction. This is a correction factor; when the reference residual mainly manifests as geometric mapping deviation, the geometric mapping parameters between the sensor coordinate system and the equipment coordinate system are corrected. ； in, and These are the geometric mapping parameters before and after the correction, respectively. The mapping correction amount is obtained based on the reference residual; when the reference residual mainly manifests as a deviation of the digital twin model, the edge parameters of the equipment digital twin model (212) are corrected: ； in, and The parameters of the edge-side equipment digital twin model before and after the correction are respectively. To equip the digital twin model with output functions, Adjust the gain matrix for the parameters; When the reference residuals mainly manifest as a slow drift in the equipment baseline response, the mean vector and covariance matrix in the equipment baseline response model parameters are corrected based on the baseline response characteristics obtained from online reference measurements. ； ； in, and These are the mean vectors of the equipment baseline response model before and after the correction; and These are the covariance matrices before and after the correction, respectively. β represents the new baseline response characteristics obtained from online reference measurements; β is the model update coefficient.

7. The intelligent monitoring system for the service status of manufacturing equipment as described in claim 1, characterized in that, The equipment baseline response normalization submodule (217) is used to establish the equipment baseline response model, assuming that the baseline response obtained during the standard calibration process is used. Baseline response characteristics of the group, the first Baseline response characteristics of the group Then the mean vector and covariance matrix of the equipment baseline response model are expressed as follows: ； ； Feature vectors acquired online Normalization is performed using the equipment baseline response model: ； in, This is the eigenvector after the baseline response is normalized.

8. The monitoring method of the intelligent monitoring system for the service status of manufacturing equipment according to any one of claims 1 to 7, characterized in that, The monitoring method includes the following steps: S1: Complete the multi-source sensor deployment, equipment communication establishment, data sensing module (100) initialization, edge computing module (200) initialization, cloud module (300) communication establishment, and in-machine self-verification benchmark module (400) initialization; S2: Perform equipment baseline response calibration by controlling the manufacturing equipment to run according to the preset no-load process parameter sequence, preset standard motion trajectory and preset reference part measurement sequence, to obtain equipment baseline response data and establish equipment baseline response model; S3: Obtain process program, controller interpolation information, phase information of process execution components and position and attitude information of equipment motion execution mechanism, and combine with equipment digital twin model (212) to predict the equipment service conditions in the current and future preset time window, and generate process semantic state sequence and manufacturing object area number; S4: Based on the process semantic state sequence, as well as the credibility and anomaly risk of the digital twin model in the previous monitoring cycle or initialization state, adaptive monitoring and scheduling are performed on the sampling rate, range, gain, channel enable status and buffer mode of the multi-source signal monitoring instrument (120), and the data storage granularity of event data. S5: The data perception module (100) performs acquisition, conditioning, spatiotemporal unified marking and pre-trigger buffering of multi-source heterogeneous signals generated during the service of manufacturing equipment according to the scheduling results determined in step S4, and the process semantic parsing submodule (213) supplements the process semantic status number and manufacturing object area number to form an event data frame; S6: The digital twin model credibility assessment submodule (214) calculates the credibility of the digital twin model for the current monitoring period based on the residual between the prediction results of the equipment digital twin model (212) and the measured values ​​of multi-source heterogeneous signals, sensor health status, spatiotemporal marker integrity and equipment baseline response offset. S7: Input the event data frame processed by the equipment baseline response normalization submodule (217) into the intelligent reasoning model (211). The intelligent reasoning model (211) obtains the status of key equipment components, process status and corresponding manufacturing object area status reasoning results based on the process semantic status number and manufacturing object area number in the event data frame, and outputs the abnormal risk of the current monitoring cycle. S8: When the credibility of the digital twin model in the current monitoring period is lower than the preset threshold or the abnormal risk continues to be higher than the preset threshold, the online self-verification submodule (216) determines the position of the on-machine self-verification benchmark module (400) or the preset verification area position of the manufacturing object according to the process semantic state number and manufacturing object area number corresponding to the abnormal event, and controls the motion actuator of the manufacturing equipment to move to the corresponding position to perform online reference measurement, and updates the sensor zero point, geometric mapping parameters, equipment baseline response model parameters and edge side equipment digital twin model parameters according to the measurement results; S9: Upload the event data frame, digital twin model credibility sequence, abnormal risk sequence, reference measurement results, equipment baseline response data, edge-side intelligent reasoning model parameters and edge-side equipment digital twin model parameters to the cloud module (300), and classify and store them by the cloud storage submodule (310); S10: When the cloud storage submodule (310) accumulates a preset number or preset period of event data, the cloud computing submodule (320) performs incremental learning update, transfer learning update or knowledge distillation update on the intelligent reasoning model (211) and equipment digital twin model (212) in the multi-source signal monitoring software (210) according to the event data, reference measurement results and edge-side intelligent reasoning model parameters and edge-side equipment digital twin model parameters, and sends the updated model parameters to the edge computing module (200) for state reasoning, digital twin prediction and adaptive monitoring scheduling in subsequent monitoring cycles.

9. The monitoring method of the intelligent monitoring system for the service status of manufacturing equipment according to claim 8, characterized in that, Step S4 includes: S41: When the current monitoring cycle is in standby mode, only the low sampling rate monitoring channel and the pre-trigger buffer unit (126) are turned on, and background data is collected at the first sampling rate; S42: When the process semantic state enters the approach segment, or when the equipment digital twin model (212) predicts that the process execution component and the manufacturing object will come into contact, energy input, material removal or surface treatment within the future preset time window, switch to the pre-trigger mode, sample at a second sampling rate higher than the first sampling rate, and retain the cached data before triggering by the pre-trigger cache unit (126). S43: When the process semantic state enters the process action establishment segment or steady-state operation segment, or when the abnormal risk exceeds the preset threshold, switch to full acquisition mode or abnormal enhanced acquisition mode, increase the sampling rate of at least one channel, and adjust the gain or range of the corresponding channel according to the channel signal amplitude, while enabling key event data storage. S44: When the process semantic state enters the process action exit segment or rollback segment, the cached data after the trigger is retained, and when the abnormal risk returns to normal and the credibility of the digital twin model is higher than the preset threshold, the standby mode is returned. The credibility of the digital twin model and the abnormal risk after returning to the standby mode are used for the adaptive monitoring scheduling of the next monitoring cycle. Step S8 includes: S81: Based on the residual type of the digital twin model, the health status of the sensor, the semantic status of the process, and the trend of abnormal risk changes, make a preliminary judgment on whether the cause of self-verification is biased towards sensor drift, deviation of the digital twin model, or abnormality of the actual service status. S82: If the judgment result is biased towards sensor drift or digital twin model deviation, the motion actuator of the control manufacturing equipment is moved to the position of the in-machine reference part (410) for reference measurement; S83: If the judgment result is biased towards the abnormal actual service status, the online self-verification submodule (216) determines the preset verification area of ​​the manufacturing object in the preset safety verification window according to the manufacturing object area number corresponding to the abnormal event, and controls the motion actuator of the manufacturing equipment to move to the preset verification area of ​​the manufacturing object to perform local verification measurement. S84: Update the sensor zero point, geometric mapping parameters, equipment baseline response model parameters, and edge-side equipment digital twin model parameters based on the reference measurement results, and recalculate the reliability and anomaly risk of the digital twin model; S85: If the risk of anomalies still exceeds the preset threshold after the update, output an alarm, speed reduction, shutdown, or compensation suggestion.

Images