A method and system for intelligent monitoring and analysis of the operating state of a detection device

Abstract

The application relates to the technical field of equipment operation monitoring, and discloses an intelligent monitoring and analysis method and system for detecting the running state of equipment, which comprises the following steps: intercepting synchronous sampling data of a multi-dimensional sensor array under a unified time reference to form a real-time test response sequence and obtaining a health state reference response sequence; performing a limited dynamic time warping operation on the two sequences to extract a warping path derivative and a path time offset; calculating the amplitude residual of an alignment node, distinguishing abnormal types according to the residual, the derivative and the offset; when it is determined that the mechanical hysteresis is within the allowable threshold range, extracting a representative time sequence deviation to generate a safety compensation amount through smooth updating and amplitude limiting processing, writing the safety compensation amount into a programmable logic controller, and adjusting the trigger time output under the condition that a preset safety constraint condition is met. The space is constructed by using multi-dimensional parameters to decouple and distinguish physical execution mechanism related abnormalities, and a smooth amplitude limiting closed loop pre-compensation mechanism is used to safely and effectively offset the mechanical hysteresis of the equipment.

Description

Technical Field

[0001] This invention relates to the field of equipment operation monitoring technology, specifically to an intelligent monitoring and analysis method and system for detecting equipment operating status. Background Technology

[0002] In automated production lines, the stable operation of inspection equipment is fundamental to ensuring production efficiency. To monitor the operating status of the underlying equipment, sensors are typically deployed near the critical actuators of the equipment to analyze its health status by collecting physical measurements during the equipment's operation.

[0003] Existing equipment status monitoring solutions typically rely on waveform amplitude threshold determination methods based on a single dimension. This monitoring mechanism struggles to effectively capture minute lags and local timing shifts caused by factors such as increased mechanical friction in actuators due to long-term operation or aging of pneumatic circuits. In practical engineering applications, existing analysis logic easily confuses this time-dimensional response lag with simple signal amplitude attenuation, leading to inaccurate anomaly type determinations.

[0004] To address mechanical hysteresis, some technical solutions attempt to extract the time offset characteristics of the response signal and use this to perform feedforward time compensation on the control commands of the equipment. However, in actual closed-loop control, due to the inherent noise in sensor signals and external transient interference in industrial environments, directly converting the extracted offset into the pre-compensation amount for the underlying actuator can easily lead to single-calculation errors directly affecting the control end, resulting in disordered fluctuations in the compensation amount and the accumulation of oscillations in the control system. Furthermore, such compensation mechanisms that directly intervene in the action time are usually independent of the equipment hardware interlocking logic. Forcibly changing the controller trigger time can easily disrupt the original timing coordination between upstream and downstream steps, leading to interference or even collision damage to related mechanical components. Therefore, existing monitoring and analysis methods still have shortcomings in decoupling and distinguishing multi-dimensional abnormal features and in ensuring the safety of feedforward compensation control. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an intelligent monitoring and analysis method and system for detecting the operating status of equipment. It solves the problem that existing equipment monitoring relies heavily on single-dimensional waveform amplitude determination, making it difficult to distinguish between minor hysteresis caused by mechanical wear and simple amplitude attenuation faults. At the same time, existing time alignment algorithms, if feedforward compensation is directly applied after extracting the time offset, are prone to causing cumulative oscillations and mechanical interference in the control system, lacking a stable and secure underlying timing pre-compensation closed-loop mechanism.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] The first aspect of this invention provides an intelligent monitoring and analysis method for the operating status of a detection device, comprising:

[0008] Using the control state switching word corresponding to the target step in the control command as the time window boundary, synchronous sampling data of the multi-dimensional sensor array is extracted under a unified time reference to generate a real-time test response sequence;

[0009] Obtain the baseline response sequence under calibrated healthy operating conditions;

[0010] Perform constrained dynamic time warping on the real-time test response sequence and the baseline response sequence to obtain the warped path, the derivative of the warped path, and the path time offset.

[0011] Calculate the amplitude residual between the real-time test response sequence and the baseline response sequence at the alignment node position corresponding to the regularized path, and distinguish the anomaly type based on the derivative of the regularized path, the path time offset, and the amplitude residual.

[0012] When the anomaly is determined to be mechanical hysteresis of the physical actuator and the amplitude residual does not exceed the amplitude residual threshold, a representative timing deviation is extracted based on the path time offset. The representative timing deviation is then smoothed, updated, and subjected to safety limiting to generate a safety compensation amount. This safety compensation amount is written into the programmable logic controller, and the trigger time of the corresponding control state switching word is adjusted under the condition of satisfying the preset safety constraints.

[0013] Furthermore, the process of extracting synchronous sampling data from a multi-dimensional sensor array under a unified time reference to form a real-time test response sequence includes: establishing a unified time reference; acquiring control command status; recording the trigger time of the control status switching word corresponding to the starting action of the target step as the start of the time window; and recording the trigger time of the control status switching word corresponding to the ending action of the target step or the starting action of the next sequential step as the end of the time window.

[0014] If the timing exceeds the preset maximum timeout threshold and the control state switching word corresponding to the end action is not obtained, the current time window boundary is forcibly truncated and a timeout abnormal alarm is output. Within the defined time window boundary, multi-channel physical measurement values ​​are synchronously extracted as synchronous sampling data at a fixed sampling frequency, and a synchronous timestamp based on a unified time reference is appended to generate a real-time test response sequence. The above process uses the controller's logic state signals to perform time-slicing of the underlying continuous sampling data, separating the valid data within the test cycle.

[0015] Furthermore, obtaining the baseline response sequence under the calibrated healthy operating state includes: continuously collecting data from multiple healthy working cycles for the same target step to generate a single-cycle response sequence; setting the trigger time of the control state switching word corresponding to the initial action of the target step in the healthy working cycle as the relative time zero point; converting the absolute synchronization timestamp in the single-cycle response sequence into a relative time offset; generating a common relative time axis; using an interpolation algorithm to resample the measurement data of the single-cycle response sequence to the corresponding node of the common relative time axis; and performing numerical fusion on all measurement data at the same relative time section to generate the baseline response sequence. This operation maps the absolute trigger time of the natural time axis to a relative coordinate system and reduces the random jitter of a single action and the sensor's background noise through interpolation and fusion operations.

[0016] Furthermore, constrained dynamic time warping is performed on the real-time test response sequence and the baseline response sequence to obtain the warped path, the derivative of the warped path, and the path time offset, including: applying a normalized weight matrix to scale-normalize each sensor channel and calculating the weighted distance as the node matching cost; setting a global constraint band within the search range to obtain the warped path, which contains warped path nodes, and the warped path nodes are represented as index coordinate pairs of corresponding sampling points; replacing the discrete index coordinates of the warped path nodes with relative timestamp records to obtain the path time coordinate points;

[0017] A fixed-width time sliding window is set up and traverses along the sequence of path time coordinates. A local least squares algorithm is used to perform linear fitting on all path time coordinates within the window to extract the normalized path derivative. The path time offset is obtained by calculating the difference between the relative time of the real-time response and the relative time of the baseline response. These operations transform the matrix element matching relationship into a time axis correspondence, quantifying the rate of change of the response process within the local window.

[0018] Furthermore, the anomaly type differentiation based on the regularized path derivative, path time offset, and amplitude residual includes: using a normalized weight matrix to eliminate dimensional differences and calculating the L2 norm as the amplitude residual; for regularized path nodes simultaneously covered by multiple time sliding windows, taking the arithmetic mean of the regularized path derivatives calculated for each of the multiple time sliding windows as the regularized path derivative at that node; and constructing a three-dimensional judgment space using the regularized path derivative, path time offset, and amplitude residual as spatial dimension coordinates for anomaly type differentiation. Utilizing these multi-dimensional parameters to construct the three-dimensional judgment space provides multi-dimensional basis for subsequent anomaly differentiation.

[0019] Furthermore, the anomaly type differentiation includes: if the amplitude residual exceeds the amplitude residual threshold, the absolute deviation of the normalized path derivative from the ideal reference value of 1 is less than the derivative deviation threshold, and the absolute value of the path time offset does not exceed the path time offset threshold, it is determined to be a response amplitude deviation anomaly; if the amplitude residual does not exceed the amplitude residual threshold, and the path time offset or the normalized path derivative meets the preset continuous deviation condition, it is determined to be a mechanical hysteresis of the physical actuator; if the amplitude residual exceeds the amplitude residual threshold, and at least one of the path time offset or the normalized path derivative exceeds the corresponding threshold range, it is determined to be a composite fault; if the sliding variance of the normalized path derivative exceeds the variance threshold, and the amplitude residual meets the preset occasional out-of-bounds condition, it is determined to be an external transient interference. This judgment logic decomposes dynamic characteristics into timing lag and amplitude maintenance parameters to separate different anomaly roots such as electrical aging, increased mechanical friction, and transient interference.

[0020] Furthermore, extracting representative time-series deviations based on path time offsets includes: obtaining a set of regular path nodes covered within a sliding window of local anomalies where mechanical lag of the physical actuator occurs; statistically filtering the set of path time offsets corresponding to the set of regular path nodes using an interquartile range algorithm to remove data points whose values ​​exceed 1.5 times the interquartile range of the upper and lower quartiles; and extracting representative path time offsets from the remaining valid data set, which are defined as representative time-series deviations for the current period. Statistical filtering algorithms are used to remove outliers, obtaining representative values ​​reflecting the actual deviations of the current period.

[0021] Furthermore, the process of smoothing and limiting the representative time series deviation to generate a safety compensation amount includes: using the representative time series deviation obtained in the current period as the residual time series deviation for the current period; adding the product of the feedforward time series compensation correction value saved in the previous working period, the smoothing update coefficient, and the residual time series deviation to update the result; and limiting the updated result to a closed interval between 0 and the preset maximum safety advance compensation amount to generate the safety compensation amount. This update mechanism can suppress drastic fluctuations in the compensation amount caused by a single data mutation, while avoiding the one-way integral accumulation of the compensation amount through the limiting process.

[0022] Furthermore, the process of writing to the programmable logic controller (PLC) and adjusting the trigger time output of the corresponding control state switching word includes: establishing a mapping table between the target step and the compensable control state switching word; only writing the safety compensation amount and adjusting the trigger time for the control state switching word marked as compensable in the mapping table; the PLC parsing the safety compensation amount value and writing the safety compensation amount into the feedforward compensation register area, control parameter area, or adjustable timer preset value storage area; shifting the level flip time corresponding to the compensable control state switching word forward on the time axis by subtracting the safety compensation amount from the standard trigger time; the final effective condition for the forward shift of the level flip time requires that the timing trigger condition be met and that the sensor corresponding to the preceding step be input with either a retreated state signal or a present state signal. These steps fine-tune the action instructions at the software level and, combined with the hardware interlocking state of the sensor, are used to offset mechanical response lag and avoid mechanical interference in the physical actuator.

[0023] A second aspect of the present invention provides an intelligent monitoring and analysis system for the operating status of testing equipment, which implements the above-mentioned intelligent monitoring and analysis method for the operating status of testing equipment, comprising:

[0024] A programmable logic controller (PLC), configured in the local execution layer of the detection equipment, is used to issue control commands according to preset process logic; a multi-dimensional sensor array is used to collect multi-channel physical measurement values ​​during the physical actuator's response to control commands; a multi-dimensional sensor acquisition module is connected to the multi-dimensional sensor array; an industrial communication bus is connected to both the PLC and the multi-dimensional sensor acquisition module to establish a unified time reference; and an edge computing unit establishes a data communication connection with the PLC and the multi-dimensional sensor acquisition module via the industrial communication bus to receive bus data streams and execute intelligent monitoring and analysis methods for operational status.

[0025] This invention provides a method and system for intelligent monitoring and analysis of the operating status of detection equipment. It has the following beneficial effects:

[0026] 1. This invention decouples the time-dimensional lag characteristics and spatial-dimensional amplitude variation characteristics of dynamic response signals by calculating the amplitude residual between the test response sequence and the reference response sequence at the alignment node, and constructing a judgment space by combining the extracted normalized path derivative and path time offset. This method effectively distinguishes between mechanical hysteresis, response amplitude deviation, complex faults, and external transient interference in physical actuators, solving the problem that traditional single-waveform amplitude determination methods easily confuse action timing lag with amplitude attenuation, and improving the accuracy of locating the cause of anomalies in underlying equipment.

[0027] 2. This invention replaces the discrete index coordinates of regular path nodes with relative timestamps, and uses local least squares and interquartile range algorithms to fit and statistically filter the coordinate points within the local sliding window. This process transforms the simple matrix array element matching relationship into a true correspondence on the physical time axis, eliminating the anomalies in difference calculations caused by identical discrete indices, and removing outlier data points in a single action. It can extract stable and representative timing deviations that reflect the true mechanical lag of the current work cycle.

[0028] 3. When mechanical lag is detected in the equipment, this invention performs residual-based smoothing and limiting processing on the extracted timing deviation to generate a safety compensation amount. Furthermore, under the condition that the hardware sensor interlocking status of the preceding steps is met, the triggering time of the controller is adjusted in advance. This underlying closed-loop pre-compensation mechanism utilizes smoothing and limiting calculations to prevent sudden changes in compensation amount and unidirectional integral accumulation caused by single data glitches. It also ensures the safety of control command execution by binding the hardware position signals of physical sensors, avoiding mechanical interference caused by forced advance of action timing. Attached Figure Description

[0029] Figure 1 This is a topology diagram of the intelligent monitoring and analysis system for the operating status of the detection equipment according to an embodiment of the present invention;

[0030] Figure 2 This is a flowchart illustrating the overall workflow of the intelligent monitoring and analysis method for the operating status of the detection equipment according to an embodiment of the present invention.

[0031] Figure 3 This is a flowchart illustrating the construction of the baseline response sequence and multi-cycle fusion in an embodiment of the present invention.

[0032] Figure 4 This is a flowchart illustrating the logic of DTW path regularization extraction and sliding window feature calculation in an embodiment of the present invention.

[0033] Figure 5 This is a flowchart of the branch decision logic for determining the type of device anomaly based on a three-dimensional judgment space in an embodiment of the present invention.

[0034] Figure 6 This is a flowchart illustrating the logic of performing feedforward timing pre-compensation and safety limiting control in an embodiment of the present invention.

[0035] Figure 7 This is a convergence logic curve diagram of the feedforward timing compensation algorithm in a continuous working cycle in an embodiment of the present invention;

[0036] Figure 8 This is a comparison of the time-domain waveforms of the fluid pipeline pressure sensor in the detection device under three states in an embodiment of the present invention. Detailed Implementation

[0037] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0038] Reference Figure 1 The present invention provides an intelligent monitoring and analysis system for the operating status of detection equipment, which may include: a programmable logic controller, a multi-dimensional sensor array, a multi-dimensional sensor acquisition module, an industrial communication bus, and an edge computing unit.

[0039] The programmable logic controller (PLC) is configured in the local execution layer of the testing equipment. The PLC issues control commands to the physical actuators of the testing equipment according to preset process logic. These control commands contain multiple control state switching words that define the start and end boundaries of the execution actions.

[0040] Multidimensional sensor arrays are deployed at key response measurement locations of the detection equipment. These arrays are used to acquire multi-channel physical measurement values ​​during the process of physical actuators responding to control commands.

[0041] The industrial communication bus connects to both the programmable logic controller (PLC) and the multidimensional sensor acquisition module. The multidimensional sensor array is connected to the multidimensional sensor acquisition module via analog input modules, digital input modules, remote I / O modules, or field data acquisition modules. The industrial communication bus is used to synchronize the system clock of the PLC with the sampling clock of the multidimensional sensor acquisition module, thereby establishing a unified time reference.

[0042] The edge computing unit establishes a data communication connection with the programmable logic controller (PLC) and the multi-dimensional sensor acquisition module via an industrial communication bus. The edge computing unit receives the bus data stream and performs logical analysis and compensation calculations for the operating status. Specifically, it receives the control state switching word output by the PLC and the synchronous sampling data output by the multi-dimensional sensor acquisition module. Based on the control state switching word, it generates a real-time test response sequence, calls the locally stored baseline response sequence, performs constrained dynamic time warping, extracts the warped path derivative, path time offset, and amplitude residual, distinguishes between anomaly types, and generates a safety compensation amount when the compensation conditions are met, then sends it to the PLC via the industrial communication bus.

[0043] Reference Figure 2 This invention provides an intelligent monitoring and analysis method for the operating status of detection equipment, comprising the following steps:

[0044] S1, the edge computing unit listens to the control instructions issued by the programmable logic controller through the industrial communication bus, uses the control state switching word corresponding to the target step as the time window boundary, and intercepts the synchronous sampling data of the multi-dimensional sensor array under a unified time reference to form the real-time test response sequence of the corresponding target step.

[0045] S2, when the detection equipment is in a manually confirmed calibrated healthy operating state, the edge computing unit performs the same data interception operation to obtain response data for multiple healthy working cycles, performs time axis alignment and numerical fusion processing on the multi-cycle response data, and generates and stores the benchmark response sequence under the standard state;

[0046] S3, the edge computing unit performs constrained dynamic time warping on the real-time test response sequence and the benchmark response sequence to obtain the warped path, converts the node coordinates corresponding to the warped path into relative timestamp coordinates and performs local linear fitting operation, extracts the derivative of the warped path, and calculates the path time offset corresponding to the warped path node.

[0047] S4, the edge computing unit calculates the amplitude residual between the real-time test response sequence and the benchmark response sequence at the alignment node position. Based on the extracted normalized path derivative, path time offset and amplitude residual, a three-dimensional judgment space is constructed. Based on the distribution of the above parameters in the three-dimensional judgment space, the anomaly type is distinguished to distinguish whether the current anomaly type of the detection device is response amplitude deviation or mechanical hysteresis of the physical actuator.

[0048] S5, when the abnormality type is determined to be mechanical hysteresis of the physical actuator and the amplitude residual does not exceed the amplitude residual threshold, the edge computing unit extracts the representative timing deviation based on the path time offset, performs smoothing update and safety limiting processing on the representative timing deviation, generates a safety compensation amount, and writes the safety compensation amount into the feedforward compensation register area of ​​the programmable logic controller through the industrial communication bus. When the programmable logic controller enters the next working cycle and is ready to execute the target step, it adjusts the trigger time output of the corresponding control state switching word according to the safety compensation amount, based on the premise of meeting the preset safety interlock conditions.

[0049] In this embodiment, the specific process of obtaining the real-time test response sequence of the corresponding target step may include the following steps:

[0050] S101, the system pre-establishes a unified time base via an industrial communication bus. Edge computing units, programmable logic controllers, and multi-dimensional sensor acquisition modules need to operate under the same time base. Specifically, the industrial communication bus can be an industrial Ethernet control automation technology bus or a process fieldbus, or other real-time communication networks. As a preferred approach, the system synchronizes the clocks of all nodes in the network through a precise time protocol within the industrial communication bus, or it can use the distributed clock synchronization mechanism of the industrial communication bus to achieve time base unification.

[0051] For the specific clock offset measurement and network latency compensation calculation principles in the precise time protocol, those skilled in the art can refer to the relevant communication standard protocols. The underlying network clock synchronization process is a well-known technology in the field and will not be elaborated upon here. Relying on a unified time base, the instruction triggering time of the programmable logic controller and the sensor sampling time can be aligned within the allowable synchronization jitter range. The specific timestamp alignment accuracy is determined by the real-time communication network used, the synchronization protocol, the controller performance, and the hardware capabilities of the acquisition module, and can achieve millisecond, sub-millisecond, or microsecond-level time resolution.

[0052] S102, after establishing a unified time base, the edge computing unit continuously monitors the message stream on the industrial communication bus, or obtains the control command status issued by the programmable logic controller (PLC) to the physical actuator by reading the open process image area, shared registers, or status variable table of the PLC. The control state switching word in the control command is specifically a status bit toggling signal at a specific storage register address within the PLC, such as the rising edge level signal at the output terminal controlling the opening of a solenoid valve or the set signal of a digital output module. The edge computing unit records the trigger time of the control state switching word corresponding to the start action of the target step as the start of the time window, and records the trigger time of the control state switching word corresponding to the end action of the target step or the start action of the next sequential step as the end of the time window.

[0053] These two logical trigger moments together constitute the target step's time window. Considering that communication packet loss or severe mechanical jamming may occur in industrial settings, preventing the termination message from being sent, the system is configured with a preset maximum timeout threshold. If, starting from the initial action trigger moment, the time exceeds the preset maximum timeout threshold without acquiring the control state switching word corresponding to the termination action, the edge computing unit forcibly truncates the current time window boundary and outputs a timeout exception alarm for that step, thereby preventing the system from entering an infinite data acquisition state.

[0054] S103, within a defined time window, the edge computing unit synchronously extracts measurement data from the multi-dimensional sensor array at a fixed sampling frequency. In typical industrial inspection equipment scenarios, to effectively capture millisecond-level mechanical motion hysteresis characteristics, this fixed sampling frequency is typically set to the range of 1kHz to 10kHz. The fixed sampling frequency can be configured based on the target step duration, the magnitude of the hysteresis to be identified, the sensor response bandwidth, and the time resolution of the programmable logic controller. The multi-dimensional sensor array may specifically include pressure transmitters, flow meters, and current transformers installed on the fluid pipeline of the inspection equipment. The edge computing unit packages the measurement values ​​of each independent sensor into a multi-channel vector, that is, it combines physical quantities of different dimensions, such as pressure, flow, and current values, collected at the same synchronous sampling time or at the same timestamp aligned cross-section into a combined data point, and adds a synchronous timestamp based on the aforementioned unified time reference to each sampled multi-channel vector.

[0055] After truncation at the time window boundaries, the edge computing unit integrates all measurement points within the window to generate a real-time test response sequence for the corresponding target step. From an engineering perspective, relying on the deterministic logic state signals of the programmable logic controller to segment the underlying acquired data can reduce the impact of non-target step data during standby, reversal, or non-target test steps on the target step response analysis. The structure of the real-time test response sequence is represented as follows:

[0056] ;

[0057] in, This represents the real-time test response sequence. This represents the total number of valid sampling points captured within the time window boundaries. Indicates the first The multidimensional sensor measurements extracted from each sampling point, and The value range is from 1 to Positive integers. Since the system has completed clock synchronization at the underlying level, the real-time test response sequence... Each multidimensional sensor measurement value All are associated with synchronized timestamp records generated based on a unified time base.

[0058] Reference Figure 3 In this embodiment, the specific process by which the edge computing unit generates and stores the baseline response sequence under standard conditions may include the following steps:

[0059] S201 defines the calibration health operating status of the testing equipment and continuously collects multi-cycle response data. The calibration health operating status typically refers to the testing equipment having just completed factory calibration, undergone in-depth maintenance and overhaul, or been confirmed by external independent instruments to be free of mechanical wear and pipeline leaks. After confirming that the testing equipment is in this operating condition, the edge computing unit, following the same instruction listening and data slicing logic as described above for acquiring the real-time test response sequence, continuously collects data for multiple health operating cycles for the same target step. To provide statistically significant representative samples during subsequent data fusion, as a preferred approach, the number of consecutively collected cycles can be set to a specific value within the range of ten to fifty. Each health operating cycle generates an independent single-cycle response sequence at this stage.

[0060] S202, performs time axis mapping and alignment processing on multi-cycle response data. Since the absolute trigger times of different health work cycles are discretely distributed on the natural time axis, the edge computing unit needs to map multiple single-cycle response sequences to a unified relative time coordinate system. In the specific computation process, the edge computing unit sets the trigger time of the control state switching word corresponding to the starting action of the target step in each health work cycle as the relative time zero point. This relative time zero point acts as the reference starting point for each cycle, used to convert the absolute synchronization timestamps attached to the sampling points in each single-cycle response sequence into relative time offsets relative to this starting point. Through this mapping mechanism, the edge computing unit enables macroscopic alignment of multiple single-cycle response sequences distributed across different absolute time periods in the relative time dimension.

[0061] S203, performs statistical numerical fusion of aligned data to generate a baseline response sequence. After completing macroscopic time dimension alignment, considering that the relative time offsets of different periods cannot be perfectly aligned due to microscopic sampling clock jitter in the industrial field, the edge computing unit pre-generates a common relative time axis with equal spacing. For each single-period response sequence, the system uses linear interpolation or spline interpolation algorithms to resample its measurement data to the corresponding node of the common relative time axis. After interpolation and resampling, for each relative time section on the common relative time axis, there are multiple multidimensional sensor measurement values ​​contributed by different working cycles.

[0062] To eliminate sensor noise floor and random mechanical jitter during a single action, the edge computing unit performs numerical fusion on all measurement data at the same relative time cross-section. Specific fusion algorithms can employ arithmetic averaging or truncated mean averaging after removing extreme points. For the engineering implementation of specific mean filtering or median filtering algorithms, those skilled in the art can configure them conventionally according to the signal-to-noise ratio requirements of the actual acquisition environment. The basic filtering processes are well-known techniques in the field and will not be elaborated upon here. After multi-cycle numerical fusion processing, the edge computing unit generates a baseline response sequence representing the standard response characteristics. The specific structure of the baseline response sequence is as follows:

[0063] ;

[0064] in, Represents the baseline response sequence. This represents the total number of valid sampling points captured and fused within the reference state time window. Indicates the first The multidimensional sensor measurements obtained by statistical fusion of multiple periods from relative time sampling points, and The value range is from 1 to A positive integer. Because... The measured values ​​in the aforementioned real-time test response sequence use the same multi-channel packaging method, and both have the same physical dimension of channel. The edge computing unit stores the constructed benchmark response sequence and its bound relative benchmark timestamp in local non-volatile memory, using it as a standard reference sequence for subsequent runtime timing deviation analysis and fault feature extraction.

[0065] Reference Figure 4 In this embodiment, the specific process by which the edge computing unit performs constrained dynamic time warping on the real-time test response sequence and the benchmark response sequence and extracts the timing deviation features may include the following steps:

[0066] S301, a constrained dynamic time warping operation is performed on the real-time test response sequence and the reference response sequence to obtain the warped path. In conventional industrial fault monitoring schemes, directly calculating the Euclidean distance between sampled waveform data usually only reflects the amplitude attenuation of the signal, and is difficult to effectively capture response timing offsets such as minor mechanical hysteresis caused by component wear. To address the above problem, this embodiment uses a constrained dynamic time warping algorithm to align the state of the real-time test response sequence and the reference response sequence. Since both the real-time test response sequence and the reference response sequence can contain multi-dimensional physical quantities with different dimensions such as pressure, flow rate, and current, before performing the constrained dynamic time warping operation, the edge computing unit uses a normalized weight matrix to perform scale normalization processing on each sensor channel, and uses the normalized weighted distance as the node matching cost in the dynamic time warping.

[0067] As an optional approach, the node matching cost can be expressed as:

[0068] ;

[0069] in, Indicates the first real-time test response sequence The sampling point and the reference response sequence The matching cost between relative time sampling points Represents the normalized weight matrix. Indicates the first Multidimensional sensor measurements extracted from each sampling point Indicates the first Multidimensional sensor measurements obtained by statistical fusion of multiple periods from relative time sampling points. This represents the L2 norm.

[0070] To limit excessive alignment path offset in alignment calculations, the system sets a global constraint band within the search range of restricted dynamic time warping.

[0071] As a preferred approach, the global constraint band can be specifically configured as a Sakoe-Chiba constraint band or an Itakura parallelogram constraint interval, with its constraint width parameter typically set to within 5% to 10% of the duration of a single target step. For the underlying dynamic programming-based cumulative distance matrix construction and backtracking solution process of the constrained dynamic time warping algorithm, those skilled in the art can refer to relevant mathematical algorithm manuals. Its basic optimal path search principle is a well-known technique in this field and will not be elaborated upon here.

[0072] Specifically, the edge computing unit can calculate the matching cost of each node within the search area defined by the global constraint band, construct a cumulative cost matrix using a dynamic programming recursive approach, and backtrack from the terminating node of the cumulative cost matrix along the direction of minimum cumulative cost to obtain the regularized path. Unlike traditional diagnostic algorithms that only utilize the scalar distance output after alignment, this technical solution obtains the regularized path generated during the alignment process. The specific structure of the regularized path is represented as follows:

[0073] ;

[0074] in, This represents the regularized path generated after aligning two sets of sequences. This represents the total number of nodes contained in the normalized path. Each node in the path is represented as an index-coordinate pair of the corresponding sampling point:

[0075] ;

[0076] in, Indicates the first step on the normalized path. The index of the sampling point corresponding to each node in the real-time test response sequence. Indicates the first step on the normalized path. The relative time sampling point index of each node in the baseline response sequence. This coordinate pair indicates that the current actual physical response test point has achieved optimal state matching with the corresponding point in the baseline response sequence.

[0077] S302 converts the node coordinates corresponding to the regularized path into relative timestamp coordinates. When a regularized path moves horizontally or vertically within a matrix, its index coordinate sequence contains consecutive identical values. This can lead to single-point division-by-zero errors during subsequent difference operations due to zero time intervals. To resolve this single-point difference calculation anomaly, the edge computing unit needs to replace the discrete index coordinates of the regularized path nodes with the relative timestamp records of the corresponding sampling points.

[0078] Before performing path normalization time coordinate mapping, the edge computing unit first converts the absolute synchronization timestamps in the real-time test response sequence into relative timestamps relative to the start time of the current target step. Specifically, if the control state switching word trigger time corresponding to the start action of the current target step is... And the first in the real-time test response sequence The absolute synchronization timestamp corresponding to each sampling point is: Then the first one in the real-time test response sequence The relative timestamp of each sampling point is defined as:

[0079] ;

[0080] The timestamps in the baseline response sequence are relative baseline timestamps established with the starting action of the same target step under healthy conditions as the zero point, denoted as . Therefore, the path nodes are regularized. The corresponding path time coordinate points are defined as follows:

[0081] ;

[0082] in, This represents the path time coordinate points after coordinate mapping. Indicates the first [number] in the real-time test response sequence The relative timestamp of each sampling point relative to the start time of the current target step. Indicates the first in the reference response sequence The relative reference timestamp of each sampling point to the start time of the reference step. This mapping mechanism transforms the simple array element matching relationship into an actual correspondence on the physical time axis, ensuring that the timing deviation analysis has a continuous physical reference.

[0083] S303, local linear fitting is performed within a preset sliding window to extract the regularized path derivative. After obtaining the path time coordinates with continuous physical time references, the edge computing unit sets a time sliding window with a fixed width on the time coordinate system and controls the time sliding window to traverse and translate along the path time coordinate sequence. As a preferred method, the fixed width is set according to the mechanical inertia time constant of the detection device, and is generally taken as tens to hundreds of sampling period widths. Within each time sliding window region, the edge computing unit uses a local least squares algorithm to perform linear fitting on all path time coordinates contained in the window to solve for the regularized path derivative representing the local time slope. The formula for calculating the regularized path derivative is as follows:

[0084] ;

[0085] in, Indicates the first The normalized path derivative extracted within a time sliding window. Indicates the first The set of path nodes covered by a time-sliding window. Indicates the first [number] in the real-time test response sequence The relative timestamp of each sampling point relative to the start time of the current target step. Indicates the first in the reference response sequence The relative reference timestamp of each sampling point relative to the start time of the reference step. Indicates the first The local statistical mean of the real-time test response sequence relative to the timestamp within a sliding time window. This represents the local statistical mean of the baseline response sequence relative to the timestamp within the sliding window at the same time. The smallest positive real number pre-configured for the system is typically set to a value of 10. -8 Up to 10 -6 Adding this minimal positive real number ensures that the denominator is never zero, thus maintaining the stability of the underlying computational program. At the physical representation level, the regularized path derivative... It reflects the rate of stretching or compression change in the response process of the detection equipment within a local time interval, thereby enabling the system to identify and locate local abnormal time sliding windows where the response timing has deviated.

[0086] S304, calculate the path time offset corresponding to the regularized path node. While extracting local slope features, the edge computing unit calculates the difference between the real-time response relative time and the baseline response relative time for each regularized path node after coordinate mapping. The calculation formula is as follows:

[0087] ;

[0088] in, Indicates the first step on the normalized path. The path time offset calculated for each node. Indicates the first [number] in the real-time test response sequence The relative timestamp of each sampling point relative to the start time of the current target step. Indicates the first in the reference response sequence The relative reference timestamp of each sampling point relative to the start time of the reference step. When When this occurs, it indicates that the actual action response of the physical actuator lags behind the calibrated healthy state, i.e., mechanical hysteresis has occurred; when This indicates that the actual response action occurred ahead of the baseline state.

[0089] This embodiment aims to extract the path time offset to quantify the actual timing deviation direction and amount within the local abnormal time sliding window, providing a basic numerical source for subsequent feedforward compensation operations.

[0090] Reference Figure 5 In this embodiment, the specific process by which the edge computing unit distinguishes anomaly types based on the extracted temporal deviation features and amplitude features may include the following steps:

[0091] S401, calculate the magnitude residual between the real-time test response sequence and the baseline response sequence at the alignment node position. Based on the regularized path node index mapping relationship obtained in the previous step, the edge computing unit extracts the first... The multidimensional sensor measurements at the sampling points, and the first sample in the baseline response sequence. The multidimensional sensor measurements are collected at several relative time sampling points. Since these measurements simultaneously contain data with different physical dimensions such as pressure, flow rate, and current, direct subtraction cannot accurately measure the overall physical deviation. To assess the overall deviation of the multidimensional measurements on a unified scale, the edge computing unit introduces a normalized weight matrix to eliminate the dimensional differences between the heterogeneous multidimensional data, and calculates the L2 norm between the two as the comprehensive amplitude residual. The formula for calculating the amplitude residual is as follows:

[0092] ;

[0093] in, Indicates the first step on the normalized path. The amplitude residuals at each node Indicates the first Multidimensional sensor measurements extracted from each sampling point Indicates the first Multidimensional sensor measurements obtained by statistical fusion of multiple periods from relative time sampling points. This represents the normalized weight matrix. As a preferred approach, the normalized weight matrix is ​​configured as a diagonal matrix, where the values ​​of its diagonal elements are the reciprocals of the full-scale values ​​of the corresponding physical sensors. By introducing this normalized weight matrix, the edge computing unit transforms the absolute deviations of multidimensional physical quantities into uniform percentage relative deviations, ensuring that the L2 norm calculation accurately reflects the degree of response amplitude deviation of the system at the current alignment cross-section.

[0094] S402 constructs a three-dimensional judgment space based on extracted parameters. In traditional industrial diagnostic systems, alarms are typically triggered by a single-dimensional waveform amplitude exceeding limit, which can easily lead to confusion about anomaly types. To accurately distinguish anomaly types, the edge computing unit uses the regularized path derivative, which represents the rate of change of local time axis stretching or compression, the path time offset, which represents macroscopic action timing deviation, and the amplitude residual, which represents response amplitude deviation, as spatial dimension coordinates to construct a three-dimensional judgment space.

[0095] Since the normalized path derivative is a feature calculated based on a local time sliding window, while the other two parameters are based on the features of a single node, to ensure tight alignment of multidimensional features on the underlying index, the edge computing unit assigns the normalized path derivative calculated within the time sliding window to all normalized path nodes covered by that window. Due to the traversal and translation characteristics of the time sliding window, the same normalized path node may be repeatedly covered by multiple adjacent time sliding windows. For nodes covered by multiple time sliding windows simultaneously, the edge computing unit takes the arithmetic mean of the normalized path derivatives calculated within each of the multiple time sliding windows as the final derivative mapping feature at that node. After alignment mapping, within this three-dimensional decision space, the amplitude residual characterizes the amplitude preservation feature of the system, while the path time offset and the normalized path derivative characterize the temporal response characteristics of the system's dynamic execution.

[0096] S403 distinguishes anomaly types based on the distribution of parameters in the three-dimensional judgment space. The edge computing unit pre-configures discrimination thresholds corresponding to the above three dimensions within the system, specifically including amplitude residual threshold, derivative deviation threshold, and path time offset threshold. As a preferred approach, the specific values ​​of these thresholds can be determined by extracting the natural statistical characteristics of the baseline health cycle test samples. For example, one or more methods can be used to determine the natural normal fluctuation boundaries of the equipment, such as mean and standard deviation, quantiles, box plot statistics, truncated statistics, or manually calibrated thresholds.

[0097] For example, the amplitude residual threshold can be determined by adding a preset multiple of the standard deviation to the mean of the amplitude residual samples within the healthy working cycle; the path time offset threshold can be determined by the preset quantile of the absolute value of the path time offset within the healthy working cycle; the derivative deviation threshold and the healthy slope range can be determined by the natural fluctuation range of the normalized path derivative relative to the ideal baseline value of 1 within the healthy working cycle; and the variance threshold can be determined by the statistical distribution of the sliding variance of the normalized path derivative within the healthy working cycle.

[0098] Preset continuous deviation conditions may include: the path time offset exceeding the path time offset threshold within multiple consecutive regular path nodes, multiple consecutive time sliding windows, or multiple consecutive working cycles; or the regular path derivative being outside the healthy slope range within multiple consecutive time sliding windows. Preset occasional out-of-bounds conditions may include: the number of nodes with amplitude residuals exceeding the amplitude residual threshold or the duration not exceeding a preset proportion or preset duration. The edge computing unit determines the landing area of ​​the parameter coordinates of each node in the three-dimensional judgment space in real time and executes the following anomaly type differentiation logic:

[0099] If amplitude residual The amplitude residual exceeds the threshold, while the absolute deviation of the normalized path derivative from the ideal reference value of 1 is less than the derivative deviation threshold, and the path time offset does not exceed the path time offset threshold. This distribution indicates that the detection equipment is basically synchronized with the calibrated healthy state in terms of time response, but the response amplitude has exceeded the limit. The edge computing unit uses this to determine that the current anomaly type belongs to the response amplitude deviation anomaly, specifically corresponding to non-time-sequential faults such as micro-leakage in the fluid pipeline of the detection equipment or natural aging of electrical components.

[0100] If amplitude residual The amplitude residual threshold was not exceeded, and the path time offset was... or regularized path derivative If the preset continuous deviation condition is met, this distribution state indicates that the response amplitude of the detection equipment is within the allowable range, but the actuator exhibits stable sluggishness, local time axis stretching, or response timing deviation in its action sequence. The edge computing unit uses this to determine that the current anomaly type belongs to mechanical hysteresis of the physical actuator, specifically corresponding to mechanical anomalies such as lack of lubrication of the cylinder guide rail or early jamming of the valve core, which increase resistance. This state serves as the criterion for entering the subsequent feedforward compensation branch.

[0101] If amplitude residual If the amplitude residual threshold is exceeded and one of the following conditions is met: the absolute value of the path time offset exceeds the path time offset threshold, or the absolute deviation of the normalized path derivative from the ideal reference value of 1 exceeds the derivative deviation threshold, or the normalized path derivative is outside the healthy slope range, this distribution indicates that both the response amplitude and the action timing have deviated beyond the limits. The edge computing unit determines that the current anomaly type is a composite fault, which usually corresponds to a severe working condition where the failure of the main air supply to the detection equipment causes a pressure drop and leads to overall mechanical lag in the pneumatic actuator. The system will trigger one or more of the following actions: shutdown interlock, alarm, re-inspection, or maintenance prompt.

[0102] As an optional decision branch, if the sliding variance calculated by the derivative of the normalized path at multiple consecutive nodes exceeds the set variance threshold, it exhibits high-frequency fluctuations, accompanied by amplitude residuals. An extremely short-lived, intermittent out-of-bounds event occurred. This distribution indicates that the system was subjected to a high-frequency sudden disturbance. The edge computing unit determined it to be external transient interference, specifically corresponding to transient voltage fluctuations in the workshop power grid or high-frequency electromagnetic crosstalk in sensor signals. The system only logs this type of interference and does not perform any control parameter correction operations. Through the above-mentioned anomaly type differentiation logic, the edge computing unit has achieved the ability to distinguish between multiple anomaly types.

[0103] Reference Figure 6 In this embodiment, the specific process by which the edge computing unit performs feedforward timing pre-compensation on the underlying controller may include the following steps:

[0104] S501, extract representative timing deviation parameters of mechanical lag in the target step. After determining that the current anomaly type belongs to mechanical lag of the physical actuator, the system needs to extract key values ​​characterizing the degree of timing deviation as the input benchmark for feedforward compensation calculation. The edge computing unit records the local anomaly time window where mechanical lag occurs as... , This includes the set of regular path nodes where the derivative of the regular path has a stable deviation or the time offset of the path has a stable excess. Edge computing unit traversal. It covers all regular path nodes and compares the path time offset calculated in the previous stage for each node.

[0105] To prevent data glitches caused by transient high-frequency pulses in the communication link, the edge computing unit uses a quartile distance algorithm before comparison. The path time offset set within the range is statistically filtered to remove data points whose values ​​exceed 1.5 times the interquartile range of the upper and lower quartiles, in order to remove outlier noise points that are outside the normal distribution range.

[0106] After outlier removal, the edge computing unit extracts the representative path time offset from the remaining valid dataset. Optionally, the representative path time offset can be obtained using the median, truncated mean, or a preset quantile. The system defines this representative path time offset as the representative time series deviation for the current period and denotes it as... .

[0107] For example, it can be calculated in the following way:

[0108] ;

[0109] Alternatively, after removing outlier points, the truncated mean of the remaining path time offsets can be calculated. The representative time series deviation of the current cycle is used to characterize the overall time series deviation of the target step in the current work cycle and serves as the input for subsequent feedforward compensation calculations.

[0110] S502 calculates the feedforward timing compensation correction value and performs safety limiting processing. To prevent oscillations in the control system caused by single acquisition errors, the edge computing unit does not directly use the representative timing deviation obtained in the current cycle as the final compensation amount, but instead performs a smooth update based on the compensation amount of the previous working cycle and the residual timing deviation of the current cycle.

[0111] The edge computing unit uses the representative timing deviation obtained in the current cycle as the residual timing deviation for the current cycle. The feedforward timing compensation correction value is updated according to the following formula:

[0112] ;

[0113] in, This represents the feedforward timing compensation correction value after the current work cycle update. This represents the feedforward timing compensation correction value saved in the previous work cycle. Indicates the smooth update coefficient. This indicates the residual timing deviation detected in the current cycle. This indicates the maximum allowable advance safety compensation amount for the current target work step. This indicates that the calculation results will be restricted to... to between.

[0114] when When this occurs, it indicates that there is still a response lag in the current cycle, and the system increases the amount of safety compensation; in subsequent tracking cycles where feedforward compensation has been enabled, if a delay is detected... This indicates that the current compensation amount has an overcompensation trend, and the system will correspondingly reduce the safety compensation amount. When When the value is less than the preset action dead zone, the system maintains the current compensation amount unchanged. This prevents the compensation amount from entering a saturation state due to unidirectional accumulation.

[0115] As an initialization process, when the detection equipment is powered on for the first time, undergoes a cold start, or has just completed a maintenance reset, the edge computing unit will... The initial value is initialized to zero. As a preferred method, the coefficients are updated smoothly. The specific value range can be set to 0.1 to 0.3 to avoid excessive changes in the compensation amount in a single instance.

[0116] In the above formula, This represents the maximum allowable safety advance compensation for the current target step. Its specific value can be determined based on the minimum dead time of the action interval, the actuator response margin, and the safety constraints of upstream and downstream steps as specified in the testing equipment's process manual. Alternatively, the maximum safety advance compensation can be the minimum value among the minimum remaining safety time of the preceding step, the allowable advance margin of the actuator, the safety interval between upstream and downstream actions, and the allowable compensation amount of the programmable logic controller's timing resolution. This ensures that the compensated trigger time does not disrupt the minimum holding time of the preceding step or the safety interval between upstream and downstream actions. The limiting function restricts the feedforward timing compensation correction value to a certain limit within the edge computing unit. to This avoids the possibility of unlimited accumulation of compensation amounts leading to mechanical impacts in the actuator or logical errors in upstream and downstream processes. The feedforward timing compensation correction value is an intermediate update amount calculated based on the compensation amount of the previous work cycle and the smoothing update coefficient. The output value after being processed by the limiting function serves as the safety compensation amount. (The last sentence appears to be incomplete and possibly refers to a different context.) Output as a safety compensation quantity.

[0117] S503 sends a compensation instruction to the programmable logic controller (PLC) to correct the triggering logic of the control state switching word. The edge computing unit encapsulates the limited safety compensation amount into a preset communication protocol message and sends it to the PLC via the industrial communication bus. After receiving the data message, the PLC parses out the corresponding safety compensation value and writes it into the feedforward compensation register area, control parameter area, or adjustable timer preset value storage area. The PLC internally has an independent adjustable timer configured for the control state switching word of this target step, or is configured with equivalent control logic that can correct the triggering time based on the safety compensation amount.

[0118] Before performing feedforward compensation, the system pre-establishes a mapping table between the target step and the control state switching word that allows compensation. This mapping table is stored in the non-volatile memory area of ​​the edge computing unit or programmable logic controller. The mapping table records at least the target step identifier, the control state switching word that allows compensation, the maximum safe advance compensation amount, the minimum hold time of the preceding step, and the safety interlock input conditions. The edge computing unit generates the safe compensation amount only for the control state switching words marked as allowing compensation in the mapping table.

[0119] When calculating the trigger time of a pre-calibrated target control state switching word that is allowed to be compensated, the programmable logic controller (PLC) shifts the level transition time of the target control state switching word forward on the time axis by subtracting the safety compensation amount from the standard trigger time, or by using an equivalent adjustable timer preset value correction method. The target control state switching word that is allowed to be compensated is one that does not disrupt the minimum hold time of the preceding step, does not affect the safety triggering conditions of the current step, and does not shorten the effective hold time of the detection. This advance of the level transition time corresponds to the safety compensation amount issued by the edge computing unit within the timing resolution and allowable synchronization jitter range of the PLC.

[0120] In industrial applications, to prevent physical collisions between the current action and the preceding action due to a simple shift in the timeline, this logic for shifting the level flip time must be based on hardware interlocking conditions. That is, the final effective condition for shifting the level flip time requires not only that the corresponding timing trigger condition be met, but also that the input of the retreated or arrived status signal from the sensor corresponding to the preceding step is simultaneously satisfied.

[0121] The aforementioned feedforward timing pre-compensation mechanism, without altering the original mechanical structure and physical actuator hardware parameters of the detection equipment, reduces the impact of physical response lag caused by cylinder guide rail wear or increased valve core friction on the timing synchronization accuracy of the target step by limiting and triggering in advance at the software control logic level. Through smooth updates via feedforward timing pre-compensation, the system enables detection equipment with slight mechanical wear to approach the target response timing under calibrated healthy conditions within the hardware-allowed time resolution range. For the address mapping configuration of the adjustable timer inside the programmable logic controller and the encoding / decoding encapsulation process of bus messages, those skilled in the art can consult the target equipment's communication control manual; its conventional instruction read / write and register modification operations are well-known technologies in the field and will not be elaborated upon here.

[0122] Taking a certain automotive engine intake manifold airtightness testing device as an example, its key testing step is the opening of the main inflation valve. The device is set to a sampling frequency of 1kHz for the multi-dimensional sensor acquisition module (i.e., the time interval between adjacent sampling points is 1ms). In a healthy operating state immediately after the testing device has completed recalibration, the edge computing unit extracts and fuses the data to generate a baseline response sequence. The total number of valid sampling points extracted and fused within the reference state time window =200 (corresponds to a 200ms intercept time window).

[0123] After six months of continuous operation, the lubricating grease inside the pneumatic directional valve dried out, increasing friction in the actuator. In the current target step, the edge computing unit monitors the bus and captures the real-time test response sequence. After performing constrained dynamic time warping, the mapped path time coordinates are... At this point, assume that the first [number] response in the real-time test response sequence... The relative timestamps of each sampling point relative to the start time of the current target step. =65ms, while the matched baseline response sequence is the first... The relative reference timestamp of each sampling point to the start time of the reference step. =50ms. According to the aforementioned formula, the first [time unit] on the normalized path... The path time offsets calculated for each node are as follows:

[0124] ;

[0125] Meanwhile, within the time sliding window (window width set to 20ms) containing this node, the edge computing unit performs a local linear fitting operation to obtain the result at the [node name missing]. Derivative of regularized path extracted within a time sliding window =1.08 (greater than the baseline value of 1, indicating that the local time axis of the response process is stretched). Additionally, the first [value] on the normalized path was calculated. Amplitude residuals at each node The value is 1.2%, which does not exceed the set amplitude residual threshold (5%). The edge computing unit makes a judgment based on the distribution of the above parameters in the three-dimensional decision space: the first on the regularized path Amplitude residuals at each node Within the normal range, but path time offset Exceeding the path time offset threshold (5ms), and the normalized path derivative The error deviates from the normal range. Based on the above judgment logic, the system determines that the anomaly type belongs to mechanical lag in the physical actuator.

[0126] After an anomaly is detected, the system enters the feedforward timing pre-compensation branch. Through median extraction and outlier removal, the residual timing deviation for the current cycle is determined. =15ms. The system is pre-configured with a smooth update coefficient. =0.2, and since the system has just enabled compensation, the feedforward timing compensation correction value saved in the previous working cycle is... =0ms, the maximum allowable safety advance compensation for the current target step. =25ms. Based on the above parameters, calculate the feedforward timing compensation correction value after the current work cycle update:

[0127] ;

[0128] The edge computing unit sends 3ms as a safety compensation amount to the programmable logic controller (PLC) via the industrial communication bus. When entering the next working cycle, the PLC outputs the trigger time of the control state switching word corresponding to the opening of the main inflation valve 3ms in advance.

[0129] To visually demonstrate the dynamic adjustment process of the above compensation logic during continuous cyclic operation of the equipment, refer to... Figure 7 , Figure 7 The horizontal axis represents the number of working cycles (times), and the vertical axis represents the time deviation and compensation amount (ms). Experimental data shows that in the first working cycle, the system detected an initial residual timing deviation of approximately 15ms (e.g., ...). Figure 7 (As shown by the solid line) and after performing the aforementioned 3ms compensation, the system, based on the above... A smoothing update coefficient of 0.2 is used to iterate continuously in subsequent cycles, ensuring that the issued safety compensation amount (such as...) is... Figure 7 The signal (shown by the dashed line) shows a smooth upward trend. After 15 consecutive working cycles, the safety compensation value stabilizes around 14.8 ms, while the residual timing deviation detected by the system converges to within the preset dead zone (±1 ms). This dynamic process indicates that the system adopts... The limiting and smoothing update mechanism avoids temporal oscillations caused by sudden changes in compensation amount.

[0130] After several cycles of smoothing compensation and stabilization, the physical execution actions achieved macroscopic alignment on the timeline. The specific alignment effect is described in [reference needed]. Figure 8 . Figure 8 The horizontal axis represents the relative timestamp (ms), and the vertical axis represents the normalized stress value (%). Figure 8 The midpoint line corresponds to the baseline health state, which reaches 50% of the normalized stress value at a relative timestamp of approximately 50 ms (e.g., Figure 8 (As shown by the solid circle).

[0131] Figure 8The dashed line corresponds to a state where hysteresis occurs and compensation logic is not enabled. Due to valve core aging, its physical pressure response is delayed, reaching 50% of the normalized pressure value only at a relative timestamp of 65ms (e.g., ...). Figure 8 (As shown by the hollow circle), this timing offset will cause the pressure holding determination time of subsequent steps to be passively shortened.

[0132] Figure 8 The solid line corresponds to the recovered state after feedforward compensation, which is the programmable logic controller based on... Figure 7 The actual sensor response waveform after the safety compensation amount is pre-triggered and stabilized is shown. It can be seen that the solid line waveform shifts to the left, reapproaching the dotted-line waveform of the calibrated health baseline state. Experimental data statistics show that after applying the compensation scheme, the root mean square time synchronization error of the device is reduced from 15.2ms before compensation to 0.8ms. This indicates that this system can utilize the feedforward timing pre-compensation logic between edge computing and the controller to correct the timing synchronization error caused by mechanical component wear without changing the physical actuator hardware.

[0133] While specific embodiments of the present invention have been described above, those skilled in the art should understand that these specific embodiments are merely illustrative. Those skilled in the art can omit, substitute, and modify the details of the above methods and systems in various ways without departing from the principles and essence of the present invention. For example, combining the above method steps to perform substantially the same function and achieve substantially the same result using substantially the same method falls within the scope of the present invention.

Claims

1. A method for intelligent monitoring and analysis of the operating status of detection equipment, characterized in that, include: Using the control state switching word of the target step in the control command as the time window boundary, synchronous sampling data of the multi-dimensional sensor array is captured under a unified time reference to generate a real-time test response sequence; Obtain the baseline response sequence under calibrated healthy operating conditions; Perform constrained dynamic time warping on the real-time test response sequence and the baseline response sequence to obtain the warped path, the derivative of the warped path, and the path time offset. Calculate the magnitude residuals of the two sequences at the alignment nodes of the regular path, and distinguish the anomaly types based on the derivative of the regular path, the path time offset, and the magnitude residuals. When the anomaly type is mechanical hysteresis of the physical actuator and the amplitude residual does not exceed the amplitude residual threshold, a representative timing deviation is extracted based on the path time offset, and a safety compensation amount is generated through smoothing update and safety limiting. This amount is written into the programmable logic controller, and the trigger time of the corresponding control state switching word is adjusted under the condition of satisfying the preset safety constraints.

2. The intelligent monitoring and analysis method for the operating status of detection equipment according to claim 1, characterized in that, The step of extracting synchronous sampling data from a multi-dimensional sensor array under a unified time reference and generating a real-time test response sequence includes: Establish a unified time standard; Acquire the control command status, record the trigger time of the control status switching word corresponding to the start action of the target step as the start of the time window, and record the trigger time of the control status switching word corresponding to the end action of the target step or the start action of the next sequential step as the end of the time window. If the timer exceeds the preset maximum timeout threshold and the control state switching word corresponding to the end action is not obtained, the current time window boundary will be forcibly truncated and a timeout error alarm will be output. Within a defined time window, multi-channel physical measurement values ​​are synchronously extracted at a fixed sampling frequency as the synchronous sampling data, and a synchronous timestamp based on the unified time reference is appended to generate the real-time test response sequence.

3. The intelligent monitoring and analysis method for the operating status of detection equipment according to claim 1, characterized in that, The acquisition of the baseline response sequence under the calibrated healthy operating state includes: Data from multiple healthy work cycles are continuously collected for the same target step to generate a single-cycle response sequence. The control state switching word trigger time corresponding to the starting action of the target step in the healthy work cycle is set as the relative time zero point, and the absolute synchronization timestamp in the single-cycle response sequence is converted into a relative time offset relative to the relative time zero point. A common relative time axis is generated based on the relative time offset. An interpolation algorithm is used to resample the measurement data of each single-cycle response sequence to the corresponding node of the common relative time axis according to the relative time offset. All measurement data on the same relative time section are numerically fused to generate the reference response sequence.

4. The intelligent monitoring and analysis method for the operating status of detection equipment according to claim 1, characterized in that, The process of performing constrained dynamic time warping on the real-time test response sequence and the baseline response sequence to obtain the warped path, the derivative of the warped path, and the path time offset includes: The scale of each sensor channel is normalized using a normalized weight matrix, and the weighted distance is calculated as the node matching cost. A global constraint band is set within the search range to obtain the regularized path. The regularized path contains regularized path nodes, and the regularized path nodes are represented as index coordinate pairs of corresponding sampling points. Replace the discrete index coordinates of the regular path nodes with relative timestamp records to obtain the path time coordinate points; A time sliding window with a fixed width is set to traverse and translate along the sequence of time coordinate points along the path. The local least squares algorithm is used to perform linear fitting operations on all path time coordinate points contained in the window to extract the derivative of the regularized path. The path time offset is obtained by calculating the difference between the real-time response relative time and the baseline response relative time.

5. The intelligent monitoring and analysis method for the operating status of the detection equipment according to claim 4, characterized in that, The distinction of anomaly types based on the normalized path derivative, the path time offset, and the amplitude residual includes: A normalized weight matrix is ​​used to eliminate dimensional differences, and the L2 norm is calculated as the magnitude residual. For a regular path node that is simultaneously covered by multiple time sliding windows, the arithmetic mean of the regular path derivatives calculated for the regular path node in each of the multiple time sliding windows is taken as the regular path derivative at the regular path node. The normalized path derivative, the path time offset, and the amplitude residual are used together as spatial dimension coordinates to construct a three-dimensional judgment space for anomaly type differentiation.

6. The intelligent monitoring and analysis method for the operating status of detection equipment according to claim 5, characterized in that, The distinction between the abnormal types includes: If the amplitude residual exceeds the amplitude residual threshold, the absolute deviation of the normalized path derivative from the ideal reference value 1 is less than the derivative deviation threshold, and the absolute value of the path time offset does not exceed the path time offset threshold, it is determined to be an anomaly of response amplitude deviation. If the amplitude residual does not exceed the amplitude residual threshold, and the path time offset or the regularized path derivative meets the preset continuous deviation condition, it is determined to be a mechanical hysteresis of the physical actuator. If the amplitude residual exceeds the amplitude residual threshold, and at least one of the path time offset or the regularized path derivative exceeds the corresponding threshold range, it is determined to be a composite fault. If the sliding variance of the normalized path derivative exceeds the variance threshold, and the amplitude residual satisfies the preset occasional out-of-bounds condition, it is determined to be an external transient disturbance.

7. The intelligent monitoring and analysis method for the operating status of detection equipment according to claim 6, characterized in that, The step of extracting representative timing deviations based on the path time offset includes: Obtain the set of regular path nodes covered within the local anomaly time sliding window where the mechanical lag of the physical actuator occurs; The interquartile range algorithm is used to statistically filter the set of path time offsets corresponding to the set of regular path nodes, and remove data points whose values ​​exceed 1.5 times the interquartile range of the upper and lower quartiles. Extract the representative path time offset from the remaining valid data set, and define it as the representative time series deviation for the current period.

8. The intelligent monitoring and analysis method for the operating status of detection equipment according to claim 7, characterized in that, The process of smoothing and updating the representative time-series deviation and performing safety-limiting processing to generate a safety compensation amount includes: The representative timing deviation obtained in the current period is determined as the residual timing deviation of the current period; The feedforward timing compensation correction value saved in the previous working cycle is added to the product of the smoothing update coefficient and the residual timing deviation to obtain the feedforward timing compensation correction value for the current cycle. The current cycle feedforward timing compensation correction value is subjected to amplitude limiting processing, which restricts it to a closed interval from 0 to a preset maximum safety advance compensation amount, and the safety compensation amount is generated and saved as the feedforward timing compensation correction value for the next working cycle.

9. The intelligent monitoring and analysis method for the operating status of detection equipment according to claim 8, characterized in that, The step of writing the safety compensation amount into the programmable logic controller and adjusting the trigger time of the corresponding control state switching word under the condition of satisfying preset safety constraints includes: Establish a mapping table between target steps and allowed compensation control state switching words, and only perform the writing of the safety compensation amount and the adjustment of the trigger time for the control state switching words marked as allowed compensation in the mapping table; The programmable logic controller parses the safety compensation value and writes the safety compensation value into the feedforward compensation register area, the control parameter area, or the adjustable timer preset value storage area. By subtracting the safety compensation amount from the standard trigger time, the time when the level flips corresponding to the allowable compensation control state switching word is shifted forward on the time axis. The final effective condition for the forward shift of the level reversal time requires that the timing trigger condition be met and that the input of the returned status signal or the arrived status signal of the corresponding sensor of the previous step be met simultaneously.

10. An intelligent monitoring and analysis system for the operating status of detection equipment, characterized in that, The method for intelligent monitoring and analysis of the operating status of the detection equipment as described in any one of claims 1-9 includes: A programmable logic controller (PLC) is configured in the local execution layer of the testing equipment to issue control commands according to preset process logic. A multi-dimensional sensor array is used to acquire multi-channel physical measurement values ​​during the process of physical actuators responding to control commands. A multi-dimensional sensor acquisition module, wherein the multi-dimensional sensor array is connected to the multi-dimensional sensor acquisition module; An industrial communication bus is connected to the programmable logic controller and the multi-dimensional sensor acquisition module respectively, for establishing a unified time reference; The edge computing unit establishes a data communication connection with the programmable logic controller and the multi-dimensional sensor acquisition module through the industrial communication bus, and is used to receive bus data streams and execute the intelligent monitoring and analysis method for the operating status of the detection equipment.

Images