Automatic equipment detection system and method based on multi-modal sensing fusion

Abstract

The invention relates to the technical field of metering calibration, in particular to an automatic equipment detection system and method based on multi-modal sensing fusion, and the system comprises an acquisition module, an abnormality determination module, an equipment determination module, an adjustment module and a correction module. According to the method, multiple types of process parameters are obtained at the same time in the detection process before a measurement calibration test, linkage association is formed in the same time sequence frame, pressure perturbation and dispersion characteristics are mapped to principal component deviation, distance judgment is conducted in combination with state vectors, meanwhile, the coupling strength is extracted through a time sequence characteristic matrix, and the measurement accuracy is improved. A correction index is calculated through a connection path and centrality of the abnormal equipment in the linkage relation chain, and a judgment threshold value is dynamically calibrated based on a coincidence rule of the abnormal equipment in a correction period; the problems of low detection precision and slow response speed in a complex large health measurement scene due to excessive dependence on single-dimension data and a deep learning model are effectively solved.

Description

Technical Field

[0001] This invention relates to the field of metrology and calibration technology, and in particular to an automated equipment testing system and method based on multimodal sensing fusion. Background Technology

[0002] With the rapid development of metrology in the health industry, various precision testing instruments are increasingly widely used in clinical settings, research, and third-party testing platforms. Over long-term use, changes in material properties, operating conditions, or the environment can cause internal calibration coefficients to shift, affecting measurement accuracy and the reliability of test results. Traditional equipment repair or calibration methods typically rely on manufacturer-sent equipment or on-site manual adjustments, which are time-consuming, inefficient, and difficult to restore the instrument to its precise operating state quickly. Therefore, rapidly and accurately acquiring the equipment's operating status before calibration or testing, identifying anomalies, and adjusting key parameters has become a significant challenge in ensuring testing accuracy and improving experimental efficiency.

[0003] Furthermore, in multi-device collaborative calibration / testing scenarios, the challenge lies not only in identifying anomalies in a single indicator, but also in the complexity of high-dimensional, multimodal signals intertwined across temporal and spatial scales. Sensors come from different manufacturers with varying ranges and sampling rates, and signals simultaneously contain short-term pulsations, slow drifts, and sudden noises. Network latency, data loss, and clock asynchrony further weaken the reliability of cross-device comparisons. The coupling between devices allows small upstream offsets to be amplified into significant downstream errors, making it difficult to distinguish between genuine faults and normal fluctuations using simple thresholds or univariate statistical methods. On the other hand, on-site calibration needs to be completed within limited time and resources, requiring the testing process to be highly sensitive to detect latent deviations, suppress false alarms to avoid unnecessary human intervention, and maintain the traceability and compliance of the results.

[0004] It is evident that existing technologies suffer from the following problems: the operating status of multiple devices and multiple parameters is difficult to synchronize and acquire with high precision, leading to anomaly identification relying on a single indicator or experience-based judgment, which may easily overlook potential deviations; the complex coupling relationship between devices makes it difficult to track the anomaly propagation path, and traditional manual calibration or static threshold methods are difficult to accurately locate the anomaly source; existing correction methods are time-consuming and inefficient, making it difficult to quickly complete parameter correction on-site, thus affecting the overall detection accuracy and experimental efficiency. Summary of the Invention

[0005] To address this, the present invention provides an automated equipment detection system and method based on multimodal perception fusion, which overcomes the problems of low detection accuracy and slow response speed in the face of complex big health measurement scenarios caused by over-reliance on single-dimensional data and deep learning models in the prior art by collecting multi-dimensional operating parameters of equipment in real time and analyzing the linkage relationship between equipment.

[0006] To achieve the above objectives, the present invention provides an automated equipment inspection system based on multimodal perception fusion, comprising: The acquisition module is used to acquire, in real time, the flow stability of the feed pipeline between the target test equipment and each associated test equipment running based on preset detection parameters during the pre-calibration test of equipment in the big health field, the pulsation amplitude of the feed pump, the vibration intensity hysteresis, and the micro pressure difference value of the linkage pressure interface, the consistency coefficient of the particle size distribution of the output material in the test linkage relationship chain, and the dispersion uniformity of the sample dispersion section, as well as the baseline drift rate of the image analysis unit in the target test equipment and the micro temperature rise gradient of the equipment operating cavity; An anomaly determination module is used to determine whether a flow path anomaly or a device performance anomaly has occurred based on the micro pressure difference value, the dispersion uniformity, and a preset anomaly determination threshold. The device determination module is used to determine a number of first abnormal associated devices based on the flow path anomaly, according to the flow stability, the pulsation amplitude and the consistency coefficient, and to determine a number of second abnormal associated devices based on the device performance anomaly, according to the vibration intensity hysteresis, the micro temperature rise gradient and the baseline drift rate. The adjustment module is used to determine a correction index based on the linkage relationship between the first abnormal associated device and the second abnormal associated device, and to adjust the preset detection parameters based on the correction index. The correction module is used to correct the preset anomaly determination threshold based on the overlap characteristics of the first abnormal associated device and the second abnormal associated device, which are re-determined within a preset correction period after adjusting the preset detection parameters.

[0007] Furthermore, the anomaly determination module is used to form a two-dimensional dataset by combining the micro-pressure difference value and the dispersion uniformity within a preset historical anomaly determination time period and perform Z-score standardization to obtain a standard dataset; and to perform principal component analysis on the standard dataset to obtain a first principal component; and to combine the micro-pressure difference value and the dispersion uniformity at the current moment into data points and calculate the squared prediction error of the data points to the first principal component; and to calculate the Hotelling statistic of the projection point of the data points to the first principal component to obtain the projection volatility; and to determine whether a flow path anomaly or a device performance anomaly has occurred based on the squared prediction error, the projection volatility, and the preset anomaly determination threshold.

[0008] Furthermore, the anomaly determination module is used to determine that the flow path anomaly has occurred when the squared prediction error is greater than the preset anomaly determination threshold and the projection fluctuation is less than or equal to the preset projection fluctuation threshold, and to determine that the equipment performance anomaly has occurred when the squared prediction error is greater than the preset anomaly determination threshold and the projection fluctuation is greater than the preset projection fluctuation threshold.

[0009] Furthermore, the device determination module is used to combine the flow stability, the pulsation amplitude, and the consistency coefficient into a three-dimensional device state vector, and calculate the Mahalanobis distance between the device state vector and the preset operating condition benchmark cluster to obtain the abnormal deviation value of the associated device, and when the abnormal deviation value is greater than the preset first abnormal deviation threshold, the corresponding associated test device is determined to be the first abnormal associated device.

[0010] Furthermore, the device determination module is used to construct a three-dimensional time-series feature matrix based on the vibration intensity hysteresis, the micro-temperature rise gradient, and the baseline drift rate within the same preset anomaly determination time period, and to calculate the singular value decomposition of the time-series feature matrix and take the maximum singular value as the system coupling strength, and to determine the corresponding associated test device as the second anomaly associated device when the system coupling strength is greater than the preset coupling strength threshold.

[0011] Furthermore, the adjustment module is used to construct an abnormal device set including all the first abnormal associated devices and the second abnormal associated devices, and, based on the test linkage relationship chain, extract the direct connection relationship and indirect connection relationship between each associated test device in the abnormal device set to form an abnormal propagation subgraph, and calculate the betweenness centrality of each node in the abnormal propagation subgraph, and calculate the correction index based on the betweenness centrality and the corresponding abnormal intensity. Wherein, the abnormal intensity is the abnormal deviation value or the system coupling strength.

[0012] Furthermore, the adjustment module is used to increase the preset feed rate in the preset detection parameters according to the relative deviation between the correction index and the maximum value of the preset calibration range when the correction index is greater than the maximum value of the preset calibration range, and to decrease the preset excitation force in the preset detection parameters according to the relative deviation between the correction index and the minimum value of the preset calibration range when the correction index is less than the minimum value of the preset calibration range.

[0013] Furthermore, the correction module is used to calculate the Jaccard similarity coefficients of all the first abnormal associated devices and the second abnormal associated devices within the preset correction period to obtain several static overlap degrees, and to calculate the mean component, trend component and fluctuation component of all static overlap degrees, and to correct the preset abnormal judgment threshold based on the mean component, trend component and fluctuation component. The trend component is calculated by performing a linear regression on all the static overlaps, and the regression coefficient is used as the trend component.

[0014] Furthermore, the correction module is used to input the mean component, trend component, and fluctuation component into a preset fuzzy inference system to obtain a comprehensive correction factor, and to correct the preset anomaly judgment threshold according to the comprehensive correction factor.

[0015] On the other hand, the present invention also provides an automated device detection method based on multimodal perception fusion, comprising: Based on a preset neural network, the flow stability of the feed pipeline between the target test equipment and each related test equipment, the pulsation amplitude of the feed pump, the vibration intensity hysteresis, and the micro pressure difference of the linkage pressure interface are acquired in real time in the big health measurement test based on preset detection parameters. The consistency coefficient of the particle size distribution of the output material in the test linkage relationship chain and the dispersion uniformity of the sample dispersion section are also acquired. The baseline drift rate of the image analysis unit in the target test equipment and the micro temperature rise gradient of the equipment operating cavity are also acquired. Based on the micro-pressure difference value, the dispersion uniformity, and the preset anomaly determination threshold, it is determined whether a flow path anomaly or a device performance anomaly has occurred. Based on the flow path anomalies, several first anomaly-related devices are determined according to the flow stability, the pulsation amplitude, and the consistency coefficient. Based on the equipment performance anomaly, several second anomaly-related devices are determined according to the vibration intensity hysteresis, the micro-temperature rise gradient, and the baseline drift rate. A correction index is determined based on the linkage relationship between the first abnormal associated device and the second abnormal associated device, and the preset detection parameters are adjusted based on the correction index. The preset anomaly determination threshold is corrected based on the overlap characteristics of the first and second abnormal associated devices, which are re-determined within a preset correction period after adjusting the preset detection parameters.

[0016] Compared with existing technologies, the beneficial effects of this invention are that by acquiring key operating parameters such as the flow stability, feed pump pulsation amplitude, vibration intensity hysteresis, micro-pressure difference, particle size distribution consistency coefficient, sample dispersion uniformity, baseline drift rate, and cavity micro-temperature gradient of the equipment in real time during the pre-calibration testing process, a comprehensive assessment of the equipment's state before the calibration test can be performed. Combining the micro-pressure difference and dispersion uniformity, abnormalities in the flow path or equipment performance can be determined. The first abnormal associated equipment is identified through flow rate, pulsation amplitude, and consistency coefficient. Simultaneously, the second abnormal associated equipment is identified based on vibration intensity hysteresis, micro-temperature gradient, and baseline drift rate. This enables the calculation of the correction index for the linkage relationship between abnormal equipment, thereby dynamically adjusting preset detection parameters and correcting the abnormality judgment threshold. This ensures that the mutual influence and coupling relationship between various operating parameters of the equipment are reasonably controlled, improving calibration accuracy and detection efficiency. At the same time, it effectively reduces the risk of misjudgment caused by equipment state fluctuations or coupling anomalies, effectively solving the problem of low detection accuracy and slow response speed when facing complex metrological scenarios in the big health field due to over-reliance on single-dimensional data and deep learning models.

[0017] Furthermore, by constructing a historical two-dimensional dataset using key parameters—micro-pressure difference and dispersion uniformity—which respectively reflect changes in the resistance of the feeding pipeline and the spatial dispersion state of the material, and through Z-score standardization and principal component analysis, the statistical principal direction jointly presented by both with changes in operating conditions can be automatically extracted, thereby capturing the inherent coupling relationship between fluid resistance fluctuations and particle dispersion behavior over time. The squared prediction error is then used to characterize the degree of anomaly in the current sample deviating from the principal direction. Combined with the amplitude of the projection fluctuation along the principal direction reflected by the Hotelling statistic, two different types of anomalies—"structural shift" and "process fluctuation"—can be monitored simultaneously. This allows flow path anomalies (such as a sudden increase in micro-pressure difference due to partial pipeline blockage) and equipment performance anomalies (such as decreased uniformity due to insufficient dispersion) to exhibit distinguishable response patterns in the statistical space. Finally, using a preset anomaly judgment threshold as a decision boundary, stable detection performance can be maintained even under noise disturbances, achieving a balance between the sensitivity of anomaly capture and the false judgment rate, significantly improving the accuracy and reliability of intelligent diagnosis in complex experimental linkage processes.

[0018] Furthermore, by setting a joint judgment mechanism of squared prediction error and projection volatility, potential flow path anomalies in the feed pipeline can be accurately identified. Squared prediction error reflects the degree of deviation of the current micro-pressure difference value and dispersion uniformity from the historical principal component direction, and can sensitively capture instantaneous deviations caused by local pipeline blockage, pump pulsation abnormalities, or flow fluctuations. Projection volatility measures the overall change amplitude along the principal component direction. When it remains low or stable, non-specific changes caused by fluctuations in the operation of the entire system can be excluded, thereby ensuring that signals deviating from the main trend truly reflect flow path anomalies. This inherent correlation between parameters enables anomaly judgment to distinguish between instantaneous local events and take into account long-term operating trends, thereby improving the accuracy, reliability, and early warning capability of flow path anomaly detection.

[0019] Furthermore, by comparing the current micro-pressure difference and dispersion uniformity of the equipment with historical standard data, the squared prediction error and projection fluctuation are calculated. When the squared prediction error exceeds the preset anomaly judgment threshold and the projection fluctuation is higher than the preset projection fluctuation threshold, the potential process performance anomalies of the equipment during the pre-calibration test can be accurately identified. This processing fully considers the inherent correlation between various parameters. For example, flow fluctuations will affect the sample dispersion uniformity, and changes in dispersion uniformity will reflect the performance deviation of local flow paths or processing sections, thereby achieving a comprehensive evaluation of the equipment status, timely detection of potential problems, and ensuring the reliability and accuracy of subsequent calibration and testing.

[0020] Furthermore, by combining flow stability, pulsation amplitude, and consistency coefficient into a three-dimensional equipment state vector and calculating its Mahalanobis distance to a preset operating condition benchmark cluster, the degree of abnormal deviation of each associated test equipment can be accurately quantified. The beneficial effects are as follows: flow stability reflects the instantaneous stability of the feed pipeline, pulsation amplitude captures the periodic fluctuations of the feed pump output, and consistency coefficient characterizes the uniformity of the output material particle size or concentration. These three factors together reflect the overall characteristics of the equipment's operating status. By considering the covariance relationship between various parameters through Mahalanobis distance, comprehensive anomalies that are difficult to detect by changes in a single parameter can be effectively identified, thereby accurately determining the first associated equipment with the anomaly. This enables early detection and quantitative assessment of system flow path deviation, pump power anomalies, and uneven material distribution, improving the automated calibration and operational stability of the test equipment.

[0021] Furthermore, by combining vibration intensity hysteresis, micro-temperature gradient, and baseline drift rate into a three-dimensional temporal feature matrix within the same time period, and extracting the maximum singular value as the system coupling strength through singular value decomposition, a quantitative characterization of the dynamic correlation between devices is achieved. Vibration intensity hysteresis reflects mechanical excitation fluctuations, micro-temperature gradient reflects the cavity thermal response, and baseline drift rate characterizes the stability of the image analysis unit. These three types of parameters interact to jointly determine the operating state of the equipment. Through the coupling strength index, system-level anomalies in the feeding and dispersing processes can be accurately identified, ensuring that the determination of the second anomaly-related equipment considers both the changes in individual parameters and the mutual coupling between multiple parameters, thereby improving the accuracy and reliability of anomaly identification.

[0022] Furthermore, by constructing a set of anomalous devices from all the first and second anomalous associated devices, and extracting the direct and indirect connections between devices based on the experimental linkage chain to form an anomalous propagation subgraph, the transmission path of the anomalous device can be clearly depicted. Further calculation of the betweenness centrality of each node and combining it with the anomalous intensity of each device yields a correction index, which quantifies the impact of the anomalous device on the overall system. Thus, the correction index not only reflects the anomalous deviation and coupling of individual devices but also comprehensively embodies the linkage relationship between devices, enabling the system to accurately identify key anomalous nodes when adjusting preset detection parameters, achieving optimized management and efficient calibration of the overall device status.

[0023] Furthermore, by dynamically adjusting the preset feed rate and preset excitation force based on the relative deviation between the correction index and the preset calibration range, real-time optimization of the experimental equipment's operating status can be achieved. When the correction index is higher than the preset upper limit, appropriately increasing the preset feed rate can compensate for deviations caused by insufficient flow path or material supply. Conversely, when the correction index is lower than the preset lower limit, reducing the preset excitation force can suppress vibration or dispersion anomalies caused by over-excitation, thereby maintaining coordination and stability among key parameters such as feeding, vibration, and dispersion. This method fully utilizes the interaction relationships between various parameters, enabling the system to adaptively adjust during dynamic operation, achieving overall performance balance and optimization, and ensuring equipment operational stability and material distribution uniformity.

[0024] Furthermore, by calculating the Jaccard similarity coefficients of the first and second associated anomalies within a preset correction period, the consistency of states among different anomaly-affected devices is quantified, thus obtaining the static overlap. The mean component, trend component, and fluctuation component are further extracted, comprehensively reflecting the average level, trend, and fluctuation amplitude of equipment anomalies. This method, based on multi-parameter historical comparison and trend analysis, allows for dynamic adjustment of the preset anomaly judgment threshold, accurately adapting to the actual performance of equipment under different operating conditions, improving the reliability of anomaly judgment, and ensuring that the linkage and coupling characteristics between parameters are fully utilized. This enables more refined monitoring and prevention of flow path anomalies and equipment performance anomalies.

[0025] Furthermore, by inputting the mean component, trend component, and fluctuation component into a preset fuzzy inference system, a comprehensive correction factor is generated. This factor is then used to dynamically adjust the preset anomaly judgment threshold, enabling the threshold to adapt to changes in equipment operating status. By comprehensively considering the average overlap, trend, and fluctuation amplitude of abnormal equipment, the corrected threshold accurately reflects the overall behavioral characteristics of the current equipment group. This effectively distinguishes between normal fluctuations and actual anomalies under different operating conditions, ensuring the accuracy and reliability of flow path anomalies and equipment performance anomalies. At the same time, it fully utilizes the synergistic relationship and change patterns between various parameters to achieve dynamic and refined management of equipment status. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of the automated equipment detection system based on multimodal perception fusion in this embodiment; Figure 2 This is the logic diagram for determining the occurrence of flow path anomalies in the anomaly determination module of this embodiment; Figure 3 This is the logic diagram for determining the occurrence of equipment performance abnormalities in the anomaly determination module of this embodiment; Figure 4 The logic diagram for determining the second abnormal associated device in the device determination module of this embodiment is shown. Detailed Implementation

[0027] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0028] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0029] Please see Figure 1As shown, this is a schematic diagram of the automated equipment detection system based on multimodal perception fusion in this embodiment. On one hand, this embodiment provides an automated equipment detection system based on multimodal perception fusion, including: The acquisition module is used to acquire, in real time, the flow stability of the feed pipeline between the target test equipment and each associated test equipment running based on preset detection parameters during the pre-calibration test of equipment in the big health field, the pulsation amplitude of the feed pump, the vibration intensity hysteresis, and the micro pressure difference value of the linkage pressure interface, the consistency coefficient of the particle size distribution of the output material in the test linkage relationship chain, and the dispersion uniformity of the sample dispersion section, as well as the baseline drift rate of the image analysis unit in the target test equipment and the micro temperature rise gradient of the equipment operating cavity; An anomaly determination module is used to determine whether a flow path anomaly or a device performance anomaly has occurred based on the micro pressure difference value, the dispersion uniformity, and a preset anomaly determination threshold. The device determination module is used to determine a number of first abnormal associated devices based on the flow path anomaly, according to the flow stability, the pulsation amplitude and the consistency coefficient, and to determine a number of second abnormal associated devices based on the device performance anomaly, according to the vibration intensity hysteresis, the micro temperature rise gradient and the baseline drift rate. The adjustment module is used to determine a correction index based on the linkage relationship between the first abnormal associated device and the second abnormal associated device, and to adjust the preset detection parameters based on the correction index. The correction module is used to correct the preset anomaly determination threshold based on the overlap characteristics of the first abnormal associated device and the second abnormal associated device, which are re-determined within a preset correction period after adjusting the preset detection parameters.

[0030] In this embodiment, the acquisition module operates in the pre-calibration test environment of equipment in the big health field, and is suitable for rapid on-site calibration and testing. The target test equipment is a combined metrology unit that performs the final material measurement and quality judgment. It includes an operating cavity, an image analysis unit, and a temperature control component, which can adjust the internal correction coefficient in real time. Each associated test device includes a feeding pipeline assembly responsible for the front-end delivery of materials, a feeding pump for quantitative feeding, a linkage pressure interface unit that constitutes pressure balance, and a dispersion device that participates in sample pre-dispersion. The acquisition module, combined with a pre-set neural network algorithm, uses different types of sensors and online detection instruments deployed on the aforementioned equipment to achieve intelligent acquisition and feature extraction of key operating conditions: a mass flow meter on the feed pipeline measures flow stability; a piezoelectric pressure sensor and pump shaft encoder at the feed pump outlet measure pulsation amplitude; a differential pressure transmitter at the linkage pressure interface provides micro-pressure differential values; an online light scattering sensor or turbidity meter at the dispersion device outlet calculates dispersion uniformity; and the consistency coefficient of the output material's particle size distribution is used to quantify whether the particle distribution of the material from the front-end feed to the final metering unit output is uniform and stable. By statistically analyzing the continuously acquired particle size sequence, such as calculating the variance or deviation of each particle size distribution and comparing it with a pre-set reference distribution, the obtained consistency coefficient can reflect the uniformity of material distribution and flow stability throughout the entire transmission and dispersion process; the image analysis unit inside the target test equipment obtains the baseline drift rate through a reference image frame combined with dark current compensation; and the thermal resistance array inside the operating chamber forms a spatial temperature field and calculates the micro-temperature gradient. All data collected from the nodes is accessed to the edge computing controller via a unified time synchronization protocol (PTP or local clock synchronization). After filtering, noise suppression, drift correction, and time alignment, a continuous multimodal data stream is formed, which is then analyzed in real time by a neural network, providing an intelligent data foundation for anomaly identification, device correlation inference, and rapid correction.

[0031] In this embodiment, the preset neural network is a multi-layer deep convolutional neural network model combined with temporal processing, specifically including: an input layer: receiving multimodal sensor data from various devices, including flow stability, pulsation amplitude, vibration intensity hysteresis, micro-pressure difference, dispersion uniformity, granularity distribution consistency, image analysis unit baseline drift rate, and cavity micro-temperature gradient; a feature extraction layer: using a convolutional neural network module to extract spatial-temporal patterns from various time series and image features, automatically capturing key fluctuations, trends, and coupling relationships; a temporal coding layer: processing the extracted feature sequences through a bidirectional long short-term memory network (Bi-LSTM) or a Transformer encoder to capture the dynamic correlation between different time steps; a fusion layer: weighted fusion of the feature vectors from various sensors to form a unified multimodal feature representation, used to comprehensively describe the operating status of the devices and their interrelationships; and an output layer: generating anomaly probabilities, performance deviation indicators, and inter-device coupling strength for each device, providing input signals for anomaly detection, device correlation analysis, and calibration adjustment. This preset neural network, after pre-training and online fine-tuning, can dynamically optimize weights according to the characteristics of different health measurement devices, enabling real-time and accurate analysis and prediction of device operating status.

[0032] In this embodiment, the preset detection parameters refer to a set of key control quantities set by the system before the operation of metrological tests in the field of health and wellness, and maintained as the operating condition benchmark. These parameters ensure that the target test equipment and related test equipment operate collaboratively under a unified test benchmark. This parameter set includes at least: a preset feed rate: the target flow rate when the feed pipeline supplies material to the feed pump, used to ensure stable feed load in the dispersion section and subsequent detection units; a preset excitation force: the target output intensity of the vibration or stirring module to maintain the material reaching the set energy level in the dispersion section; a preset pressure stability range: the allowable range of micro-pressure difference at the linkage pressure interface, used to limit flow path blockage or local pressure fluctuations; a preset dispersion energy density: the target power of the dispersion section to achieve the set dispersion uniformity; a preset image acquisition benchmark: the target exposure, sampling frame rate, and allowable baseline drift threshold of the image analysis unit; and a preset temperature control uniformity range: the allowable temperature gradient range of the equipment operating cavity. These preset detection parameters can be input by the test user through the interface or automatically loaded according to the standard test protocol, and serve as the control targets and anomaly evaluation benchmarks for each functional device during the test operation.

[0033] In this embodiment, the experimental linkage chain refers to a dynamic operational structure in metrological testing in the field of health and wellness, composed of a target testing device and multiple related testing devices, arranged according to material flow direction and functional dependencies. It serves as a directed link model describing the causal coupling and data dependencies between the devices. The linkage chain begins with the front-end material supply and includes: the transfer of material flow from the supply pipeline to the feed pump; the pulsating output of the feed pump directly affecting the instantaneous flow rate entering the dispersion device; the dispersion uniformity of the dispersion device determining the initial distribution state of the sample before entering the detection unit; vibration intensity and dispersion uniformity jointly affecting the final material state and measurement results; the output material data being collected by the image analysis unit and fed back to the target testing device for determining the overall detection effect; and the linkage pressure interface providing pressure coupling information between the pipeline and the dispersion section for judging flow path blockage or local load changes. The above-mentioned linkage chain is modeled using an internal system topology definition file (e.g., JSON or graph database format), explicitly representing the input variables, output variables, and the transmission method of their influence on downstream devices for each device node. During the anomaly identification and equipment correlation analysis phase, the system determines the anomaly propagation path based on the linkage chain, identifies which equipment may generate anomalies due to upstream deviations, and generates an anomaly propagation subgraph accordingly, providing a structural basis for subsequent automatic calibration and threshold correction.

[0034] In this embodiment, the entire data processing and analysis process is completed under a cloud-edge computing collaborative architecture. Various sensor parameters collected by the target experimental equipment and related experimental equipment are first preprocessed and initially calculated in real time at the edge node, such as Z-score normalization, temporal feature matrix construction, and local anomaly detection. Subsequently, the processed feature data is uploaded to the cloud, where the more powerful computing resources are utilized for principal component analysis, singular value decomposition, betweenness centrality calculation, and fuzzy inference processing, ultimately generating correction exponents and correction factors. Through cloud-edge collaborative computing, not only is real-time performance and high accuracy of the calculations guaranteed, but historical data and global information are also fully utilized to achieve dynamic monitoring and intelligent calibration of the equipment status.

[0035] The preset anomaly judgment threshold is a statistical judgment boundary used by the system to distinguish between normal operation and abnormal conditions. It depends on the historical distribution of micro-pressure difference values, the level of dispersion uniformity fluctuation, and the stability range of particle size consistency indicators of the target test equipment and its associated test equipment under steady-state conditions. This threshold is typically set within the statistical significance range corresponding to 3 to 5 standard deviations. In this embodiment, a boundary value corresponding to 3.5 standard deviations is used, which ensures a low false alarm rate while maintaining sufficient sensitivity to early deviations such as sudden changes in flow path load or equipment performance degradation. The preset correction cycle refers to the fixed duration for automatically correcting the offset, drift, or cumulative errors generated by the target test equipment and its associated equipment during continuous operation. It depends on the stability level of the equipment, material adhesion, and the stability of the preset excitation force and preset feed rate, and is typically set to 30 to 120 minutes. In this embodiment, it is set to 60 minutes, which can promptly eliminate systematic deviations without excessively increasing the correction frequency, ensuring that the measurement data across the entire chain remains stable and accurate over the long term.

[0036] By acquiring key operating parameters such as flow stability, feed pump pulsation amplitude, vibration intensity hysteresis, micro-pressure difference, particle size distribution consistency coefficient, sample dispersion uniformity, baseline drift rate, and cavity micro-temperature gradient in real time during the pre-calibration testing process, the system can comprehensively assess the equipment's condition before calibration. Combining micro-pressure difference and dispersion uniformity, it can identify flow path or equipment performance anomalies. The system identifies the first associated anomaly based on flow rate, pulsation amplitude, and consistency coefficient, and identifies the second associated anomaly based on vibration intensity hysteresis, micro-temperature gradient, and baseline drift rate. This allows for the calculation of a correction index for the linkage between abnormal devices, dynamically adjusting preset detection parameters and correcting anomaly judgment thresholds. This ensures reasonable control of the mutual influence and coupling relationships between various equipment operating parameters, improving calibration accuracy and detection efficiency. Simultaneously, it effectively reduces the risk of misjudgment caused by equipment status fluctuations or coupling anomalies, effectively solving the problem of low detection accuracy and slow response speed when facing complex metrological scenarios in the big health field due to over-reliance on single-dimensional data and deep learning models.

[0037] Specifically, the anomaly determination module is used to form a two-dimensional dataset by combining the micro-pressure difference value and the dispersion uniformity within a preset historical anomaly determination period and perform Z-score standardization to obtain a standard dataset; and to perform principal component analysis on the standard dataset to obtain a first principal component; and to combine the micro-pressure difference value and the dispersion uniformity at the current moment into data points and calculate the squared prediction error of the data points to the first principal component; and to calculate the Hotelling statistic of the projection point of the data points to the first principal component to obtain the projection volatility; and to determine whether a flow path anomaly or a device performance anomaly has occurred based on the squared prediction error, the projection volatility, and the preset anomaly determination threshold.

[0038] In this embodiment, the calculation of Z-score standardization, principal component analysis, squared prediction error, and Hotelling statistic are all conventional statistical methods widely used in industrial process monitoring and data mining, with complete mathematical definitions and mature software implementations, and will not be elaborated further.

[0039] By constructing a historical two-dimensional dataset using key parameters—micro-pressure difference and dispersion uniformity—which respectively reflect changes in the resistance of the feed pipeline and the spatial dispersion state of the material, and through Z-score standardization and principal component analysis, the statistical principal direction jointly presented by both with changes in operating conditions can be automatically extracted. This captures the inherent coupling relationship between fluid resistance fluctuations and particle dispersion behavior over time. Furthermore, the squared prediction error is used to characterize the degree of anomaly in the current sample's deviation from the principal direction. Combined with the amplitude of the projection fluctuation along the principal direction reflected by the Hotelling statistic, two different types of anomalies—"structural shift" and "process fluctuation"—can be monitored simultaneously. This allows flow path anomalies (such as a sudden increase in micro-pressure difference due to partial pipeline blockage) and equipment performance anomalies (such as decreased uniformity due to insufficient dispersion) to exhibit distinguishable response patterns in the statistical space. Finally, using a preset anomaly judgment threshold as a decision boundary, stable detection performance can be maintained even under noise disturbances, achieving a balance between the sensitivity of anomaly capture and the false judgment rate, significantly improving the accuracy and reliability of intelligent diagnosis in complex experimental linkage processes.

[0040] Please see Figure 2 As shown, it is the logic diagram for determining the occurrence of flow path anomalies by the anomaly determination module in this embodiment. In this embodiment, the anomaly determination module is used to determine the occurrence of the flow path anomaly when the squared prediction error is greater than the preset anomaly determination threshold and the projection fluctuation is less than or equal to the preset projection fluctuation threshold.

[0041] The preset projection fluctuation threshold is a boundary value used to judge the fluctuation amplitude of the current sample along the principal component direction. It depends on the normal fluctuation range of historical running data and the stability of system operation. It is usually set between 0.01 and 0.05. In this embodiment, it is set to 0.03, which can effectively distinguish between the overall fluctuation of the system and the local flow path anomaly, and ensure the sensitivity and accuracy of anomaly judgment.

[0042] By establishing a joint judgment mechanism of squared prediction error and projection volatility, potential flow path anomalies in the feed pipeline can be accurately identified. Squared prediction error reflects the degree of deviation of the current micro-pressure difference value and dispersion uniformity from the historical principal component direction, and can sensitively capture instantaneous deviations caused by local pipeline blockage, pump pulsation abnormalities, or flow fluctuations. Projection volatility measures the overall change amplitude along the principal component direction. When it remains low or stable, non-specific changes caused by fluctuations in the operation of the entire system can be excluded, thereby ensuring that signals deviating from the main trend truly reflect flow path anomalies. This inherent correlation between parameters enables anomaly judgment to distinguish between instantaneous local events and take into account long-term operating trends, thereby improving the accuracy, reliability, and early warning capability of flow path anomaly detection.

[0043] Please see Figure 3 As shown, it is a logic diagram for determining the occurrence of equipment performance abnormality by the anomaly determination module in this embodiment. In this embodiment, the anomaly determination module is used to determine that the equipment performance abnormality has occurred when the squared prediction error is greater than the preset anomaly determination threshold and the projection fluctuation is greater than the preset projection fluctuation threshold.

[0044] By comparing the current micro-pressure difference and dispersion uniformity of the equipment with historical standard data, the squared prediction error and projection fluctuation are calculated. When the squared prediction error exceeds the preset anomaly judgment threshold and the projection fluctuation is higher than the preset projection fluctuation threshold, the system can accurately identify possible process performance anomalies in the equipment during the pre-calibration test. This process fully considers the inherent correlation between various parameters. For example, flow fluctuations can affect the sample dispersion uniformity, and changes in dispersion uniformity can reflect performance deviations in local flow paths or processing sections. This enables a comprehensive assessment of the equipment status, timely detection of potential problems, and ensures the reliability and accuracy of subsequent calibration and testing.

[0045] Specifically, the device determination module is used to combine the flow stability, the pulsation amplitude and the consistency coefficient into a three-dimensional device state vector, and calculate the Mahalanobis distance between the device state vector and the preset operating condition benchmark cluster to obtain the abnormal deviation value of the associated device. When the abnormal deviation value is greater than the preset first abnormal deviation threshold, the corresponding associated test device is determined to be the first abnormal associated device.

[0046] In this embodiment, the flow stability, pulsation amplitude, and consistency coefficient are combined into a three-dimensional device state vector, represented as: xt=[Ft, Pt, Ct], where xt is the device state vector, Ft is the flow stability, Pt is the pulsation amplitude, and Ct is the consistency coefficient.

[0047] In this embodiment, the Mahalanobis distance between the computing device state vector and the preset operating condition benchmark cluster is expressed as: , where DM is the Mahalanobis distance, i.e., the abnormal deviation value, μ is the mean vector of the preset working condition benchmark cluster, and Σ is the covariance matrix of the preset working condition benchmark cluster.

[0048] In this embodiment, the method for constructing the preset operating condition benchmark cluster is as follows: based on historical operating data consisting of the flow stability, pulsation amplitude, and consistency coefficient collected by the device during historical, fault-free, and stable operation; the historical operating data is preprocessed to remove outliers caused by obvious sensor interference or communication anomalies, ensuring the validity of the data; the preset operating condition benchmark cluster is characterized by the statistical characteristics of the preprocessed historical operating data, specifically including: mean vector μ: a three-dimensional vector whose three components are the arithmetic mean of the flow stability, pulsation amplitude, and consistency coefficient in all historical data, used to characterize the central trend of normal system operation; covariance matrix Σ: a 3x3 symmetric matrix whose diagonal elements are the variances of the flow stability, pulsation amplitude, and consistency coefficient, used to characterize the normal fluctuation range of each parameter; its off-diagonal elements are the covariances between corresponding parameters, used to characterize the linkage relationship and coupling degree between parameters under normal operating conditions. The preset operating condition benchmark cluster, through the mean vector and covariance matrix, fully defines the multi-parameter joint probability distribution characteristics of the equipment under normal operating conditions, providing a reliable mathematical benchmark for subsequent calculation of abnormal deviations based on Mahalanobis distance.

[0049] By combining flow stability, pulsation amplitude, and consistency coefficient into a three-dimensional equipment state vector and calculating its Mahalanobis distance to a preset operating condition benchmark cluster, the degree of abnormal deviation of each associated test equipment can be accurately quantified. Flow stability reflects the instantaneous stability of the feed pipeline, pulsation amplitude captures the periodic fluctuations of the feed pump output, and consistency coefficient characterizes the uniformity of the output material particle size or concentration. These three factors together reflect the overall characteristics of the equipment's operating status. By considering the covariance relationship between various parameters through Mahalanobis distance, comprehensive anomalies that are difficult to detect by changes in a single parameter can be effectively identified, thereby accurately determining the first associated equipment with the anomaly. This enables early detection and quantitative assessment of system flow path deviation, pump power anomalies, and uneven material distribution, improving the automated calibration and operational stability of test equipment.

[0050] Please see Figure 4As shown, this is the determination logic diagram of the device determination module in this embodiment for determining the second abnormal associated device. In this embodiment, the device determination module is used to construct a three-dimensional time-series feature matrix based on the vibration intensity hysteresis, the micro-temperature rise gradient, and the baseline drift rate within the same preset abnormal determination time. It also calculates the singular value decomposition of the time-series feature matrix and takes the maximum singular value as the system coupling strength. When the system coupling strength is greater than the preset coupling strength threshold, it determines that the corresponding associated test device is the second abnormal associated device.

[0051] In this embodiment, the three-dimensional temporal feature matrix is ​​constructed as follows: A = [ΔT{tk}, ..., ΔTt; G{tk}, ..., Gt; D{tk}, ..., Dt], where A is a (k+1)×3-dimensional temporal feature matrix, ΔT{tk} to ΔTt is the sequence of vibration intensity hysteresis from time tk to t, G{tk} to Gt is the sequence of the corresponding micro-temperature gradient, and D{tk} to Dt is the sequence of the corresponding baseline drift rate.

[0052] In this embodiment, the system coupling strength is calculated as: σmax=max[SVD(A)], where σmax is the maximum singular value obtained after the time series feature matrix is ​​decomposed into singular values, i.e., the system coupling strength.

[0053] By combining vibration intensity hysteresis, micro-temperature gradient, and baseline drift rate into a three-dimensional temporal feature matrix within the same time period, and extracting the maximum singular value as the system coupling strength through singular value decomposition, a quantitative characterization of the dynamic correlation between devices is achieved. Vibration intensity hysteresis reflects mechanical excitation fluctuations, micro-temperature gradient reflects the cavity thermal response, and baseline drift rate characterizes the stability of the image analysis unit. These three types of parameters interact to jointly determine the operating state of the equipment. Through the coupling strength index, system-level anomalies in the feeding and dispersing processes can be accurately identified, ensuring that the determination of the second anomaly-related equipment considers both the changes in individual parameters and the mutual coupling between multiple parameters, thereby improving the accuracy and reliability of anomaly identification.

[0054] Specifically, the adjustment module is used to construct an abnormal device set including all the first abnormal associated devices and the second abnormal associated devices, and, based on the test linkage relationship chain, extract the direct connection relationship and indirect connection relationship between each associated test device in the abnormal device set to form an abnormal propagation subgraph, and calculate the betweenness centrality of each node in the abnormal propagation subgraph, and calculate the correction index based on the betweenness centrality and the corresponding abnormal intensity. Wherein, the abnormal intensity is the abnormal deviation value or the system coupling strength.

[0055] In this embodiment, the correction index calculated based on the betweenness centrality and the corresponding anomaly strength is expressed as follows: Where CI is the correction index, Bi is the betweenness centrality of the i-th anomalous device in the set of anomalous devices, Ai is the anomalous strength of the i-th anomalous device in the set of anomalous devices (the anomalous deviation value for the first anomalous associated device and the system coupling strength for the second anomalous associated device), and N is the total number of anomalous devices.

[0056] In this embodiment, constructing a set of anomalous devices is a conventional data aggregation operation; extracting device connection relationships based on existing experimental linkage chains to construct network subgraphs is a standard practice in graph theory and complex network analysis; the calculation of node betweenness centrality uses the classic Brandes algorithm, a common technique for network centrality analysis; and the calculation of the comprehensive index based on node centrality and anomalous strength can be directly implemented using a weighted summation mathematical model. The preset anomalous deviation weight corresponding to the anomalous deviation value is used to measure the contribution of a single device's state deviation to the overall correction index. This weight depends on the typical magnitude of each device's deviation in historical operating data and the system's sensitivity, and is typically set between 0.1 and 1.0. In this embodiment, it is set to 0.6, which reasonably reflects the impact of device anomalousness on the overall system. The preset coupling weight corresponding to the system coupling strength is used to quantify the impact of inter-device coupling anomalies on the correction index. This weight depends on the linkage strength between devices and the overall system stability, and is typically set between 0.1 and 1.0. In this embodiment, it is set to 0.8, which highlights the guiding role of key coupling anomalies in overall adjustment.

[0057] By constructing a set of anomalous devices from all devices associated with the first and second anomalies, and extracting the direct and indirect connections between devices based on the experimental linkage chain to form an anomaly propagation subgraph, the transmission path of anomalies between devices can be clearly depicted. Furthermore, by calculating the betweenness centrality of each node and combining it with the anomaly intensity of each device to obtain a correction index, the impact of anomalies on the overall system can be quantified. Thus, the correction index not only reflects the anomaly deviation and coupling of individual devices but also comprehensively embodies the linkage relationships between devices, enabling the system to accurately identify key anomaly nodes when adjusting preset detection parameters, achieving optimized management and efficient calibration of the overall device status.

[0058] Specifically, the adjustment module is used to increase the preset feed rate in the preset detection parameters according to the relative deviation between the correction index and the maximum value of the preset calibration range when the correction index is greater than the maximum value of the preset calibration range, and to decrease the preset excitation force in the preset detection parameters according to the relative deviation between the correction index and the minimum value of the preset calibration range when the correction index is less than the minimum value of the preset calibration range.

[0059] Where L'=L×(1+k1×︱Y-Ymax︱ / Ymax)U'=U×(1-k2×︱Y-Ymin︱ / Ymin), L' is the adjusted preset feed rate, L is the original preset feed rate, k1 is the preset feed adjustment coefficient, Y is the correction index, Ymax is the maximum value of the preset calibration range, U' is the adjusted preset excitation force, U is the original preset excitation force, k2 is the preset excitation adjustment coefficient, and Ymin is the minimum value of the preset calibration range.

[0060] The maximum value of the preset calibration range is the upper limit of the correction index allowed by the system, which depends on the fluctuation of the correction index during the historical stable operation of the equipment. It is usually set between 80% and 120%, and in this embodiment, it is set to 110%, which can reasonably adjust the correction index if it is too high while ensuring the safety and stability of the equipment. The minimum value of the preset calibration range is the lower limit of the correction index allowed by the system, which depends on the fluctuation of the correction index during the historical stable operation of the equipment. It is usually set between 60% and 90%, and in this embodiment, it is set to 70%, which can effectively suppress the over-response of the equipment when the correction index is too low. The preset feeding rate is the initial material conveying rate of the target test equipment, which depends on the test process requirements and equipment specifications. It is usually set between 5 and 50 mL / min, and in this embodiment, it is set to 20 mL / min. The feed rate is set at mL / min, ensuring a stable material supply without causing abnormal vibration. The preset feed adjustment coefficient is a proportional coefficient used to calculate the dynamic adjustment range of the feed rate. It depends on the sensitivity of the correction exponent and the equipment response characteristics, and is usually set between 0.01 and 0.2. In this embodiment, it is set to 0.05, which enables smooth and controllable feed adjustment. The preset excitation force is the initial excitation input intensity of the test equipment. It depends on the material characteristics and dispersion requirements, and is usually set between 10 and 100 N. In this embodiment, it is set to 50 N, which provides sufficient vibration force to promote uniform material dispersion. The preset excitation adjustment coefficient is a proportional coefficient used to calculate the dynamic adjustment range of the excitation force. It depends on the sensitivity of the correction exponent and the equipment vibration response, and is usually set between 0.01 and 0.2. In this embodiment, it is set to 0.05, which can ensure the vibration effect while preventing excessive vibration from causing uneven dispersion or equipment damage.

[0061] By dynamically adjusting the preset feed rate and preset excitation force based on the relative deviation between the correction index and the preset calibration range, real-time optimization of the experimental equipment's operating status can be achieved. When the correction index is higher than the preset upper limit, appropriately increasing the preset feed rate can compensate for deviations caused by insufficient flow path or material supply. Conversely, when the correction index is lower than the preset lower limit, decreasing the preset excitation force can suppress vibration or dispersion anomalies caused by over-excitation, thereby maintaining coordination and stability among key parameters such as feeding, vibration, and dispersion. This method fully utilizes the interaction relationships between various parameters, enabling the system to adaptively adjust during dynamic operation, achieving overall performance balance and optimization, and ensuring equipment operational stability and material distribution uniformity.

[0062] Specifically, the correction module is used to calculate the Jaccard similarity coefficients of all the first abnormal associated devices and the second abnormal associated devices within the preset correction period to obtain several static overlap degrees, and to calculate the mean component, trend component and fluctuation component of all static overlap degrees, and to correct the preset abnormal judgment threshold based on the mean component, trend component and fluctuation component. The trend component is calculated by performing a linear regression on all the static overlaps, and the regression coefficient is used as the trend component.

[0063] Where Ji = |Ei∩Pi| / |Ei∪Pi|, Ji is the Jaccard similarity coefficient at the i-th time within the preset correction period, Ei is all the first abnormal associated devices at the i-th time, and Pi is all the second abnormal associated devices at the i-th time.

[0064] By calculating the Jaccard similarity coefficients of the first and second associated anomalies within a preset correction period, the consistency of states among different anomalous devices is quantified, thus obtaining the static overlap. Furthermore, mean, trend, and fluctuation components are extracted, comprehensively reflecting the average level, trend, and fluctuation amplitude of equipment anomalies. This method, based on multi-parameter historical comparison and trend analysis, allows for dynamic adjustment of the preset anomaly judgment threshold, accurately adapting to the actual performance of equipment under different operating conditions, improving the reliability of anomaly judgment, and ensuring that the linkage and coupling characteristics between parameters are fully utilized. This enables more refined monitoring and prevention of flow path anomalies and equipment performance anomalies.

[0065] Specifically, the correction module is used to input the mean component, trend component, and fluctuation component into a preset fuzzy inference system to obtain a comprehensive correction factor, and to correct the preset anomaly judgment threshold according to the comprehensive correction factor.

[0066] Where Q'=Q×exp(λ×(R-R0) / R0), Q' is the corrected preset anomaly detection threshold, Q is the original preset anomaly detection threshold, λ is the comprehensive correction factor, R is the mean component, and R0 is the preset benchmark overlap rate.

[0067] In this embodiment, the pre-defined fuzzy inference system is constructed based on mature fuzzy logic theory. Its system architecture and operating mechanism are as follows: The system uses the mean component, trend component, and fluctuation component as input variables. Each input variable is divided into three fuzzy levels: "low," "medium," and "high," and the transformation from precise values ​​to fuzzy concepts is achieved through a triangular membership function. The system output variable is a comprehensive correction factor, divided into three fuzzy levels: "negative adjustment," "maintain," and "positive adjustment." The system kernel establishes a rule base containing 27 fuzzy rules, each rule describing the nonlinear mapping relationship between input and output in "IF-THEN" form. The system employs a Mamdani-type inference mechanism, performing fuzzy inference through maximum-minimum value synthesis operations, and finally using a centroid-based defuzzifier to transform the output fuzzy set into a precise comprehensive correction factor. This fuzzy inference system, by transforming expert experience into computable fuzzy rules, achieves intelligent judgment and decision output for complex system states.

[0068] By inputting the mean component, trend component, and fluctuation component into a preset fuzzy inference system, a comprehensive correction factor is generated. This factor is then used to dynamically adjust the preset anomaly judgment threshold, enabling the threshold to adapt to changes in equipment operating status. The system comprehensively considers the average overlap, trend, and fluctuation amplitude of abnormal equipment, allowing the corrected threshold to accurately reflect the overall behavioral characteristics of the current equipment group. This effectively distinguishes between normal fluctuations and actual anomalies under different operating conditions, ensuring the accuracy and reliability of flow path anomalies and equipment performance anomalies. At the same time, it fully utilizes the synergistic relationship and change patterns between various parameters to achieve dynamic and refined management of equipment status.

[0069] On the other hand, this embodiment also provides an automated device detection method based on multimodal perception fusion, including: Based on a preset neural network, the flow stability of the feed pipeline between the target test equipment and each related test equipment, the pulsation amplitude of the feed pump, the vibration intensity hysteresis, and the micro pressure difference of the linkage pressure interface are acquired in real time in the big health measurement test based on preset detection parameters. The consistency coefficient of the particle size distribution of the output material in the test linkage relationship chain and the dispersion uniformity of the sample dispersion section are also acquired. The baseline drift rate of the image analysis unit in the target test equipment and the micro temperature rise gradient of the equipment operating cavity are also acquired. Based on the micro-pressure difference value, the dispersion uniformity, and the preset anomaly determination threshold, it is determined whether a flow path anomaly or a device performance anomaly has occurred. Based on the flow path anomalies, several first anomaly-related devices are determined according to the flow stability, the pulsation amplitude, and the consistency coefficient. Based on the equipment performance anomaly, several second anomaly-related devices are determined according to the vibration intensity hysteresis, the micro-temperature rise gradient, and the baseline drift rate. A correction index is determined based on the linkage relationship between the first abnormal associated device and the second abnormal associated device, and the preset detection parameters are adjusted based on the correction index. The preset anomaly determination threshold is corrected based on the overlap characteristics of the first and second abnormal associated devices, which are re-determined within a preset correction period after adjusting the preset detection parameters.

[0070] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A device automation detection system based on multi-modal perception fusion, characterized in that, The method comprises the following steps: an acquisition module is configured to acquire, in real time, a flow stability of a feed pipe between a target test device and each associated test device, a pulsation amplitude of a feed pump, a vibration intensity return difference, a micro-pressure difference value of a linkage pressure interface, a consistency coefficient of a particle size distribution of an output material in a test linkage relationship chain, a dispersion uniformity of a sample dispersion section, a baseline drift rate of an image analysis unit in the target test device, and a micro-temperature rise gradient of a device operation cavity, based on a preset neural network during a pre-test detection process of a calibration test of a device in a large health field; an anomaly determination module is configured to determine a flow path type anomaly or a device performance anomaly according to the micro-pressure difference value, the dispersion uniformity, and a preset anomaly determination threshold value; a device determination module is configured to determine a plurality of first anomaly-associated devices according to the flow stability, the pulsation amplitude, and the consistency coefficient based on the flow path type anomaly, and to determine a plurality of second anomaly-associated devices according to the vibration intensity return difference, the micro-temperature rise gradient, and the baseline drift rate based on the device performance anomaly; an adjustment module is configured to determine a correction index according to a linkage relationship of the first anomaly-associated devices and the second anomaly-associated devices, and to adjust the preset detection parameter according to the correction index; a correction module is configured to correct the preset anomaly determination threshold value according to a coincidence feature of the first anomaly-associated devices and the second anomaly-associated devices that are re-determined within a preset correction period after the preset detection parameter is adjusted.

2. The multi-modal perception fusion based equipment automation detection system according to claim 1, wherein, The anomaly determination module is configured to combine the micro-pressure difference value and the dispersion uniformity within a preset anomaly determination time period to form a two-dimensional data set, and to perform Z-score standardization to obtain a standard data set, to perform principal component analysis on the standard data set to obtain a first principal component, to combine the micro-pressure difference value and the dispersion uniformity at a current time to form a data point, to calculate a squared prediction error of the data point to the first principal component, to calculate a Hotteling statistic of a projection point of the data point in the first principal component to obtain a projection fluctuation degree, and to determine a flow path type anomaly or a device performance anomaly according to the squared prediction error, the projection fluctuation degree, and the preset anomaly determination threshold value.

3. The device automation detection system based on multi-modal perception fusion according to claim 2, characterized in that, The anomaly determination module is configured to determine the flow path type anomaly when the squared prediction error is greater than the preset anomaly determination threshold value and the projection fluctuation degree is less than or equal to a preset projection fluctuation threshold value, and to determine the device performance anomaly when the squared prediction error is greater than the preset anomaly determination threshold value and the projection fluctuation degree is greater than the preset projection fluctuation threshold value.

4. The device automation detection system based on multi-modal perception fusion of claim 3, wherein, The device determination module is configured to combine the flow stability, the pulsation amplitude, and the consistency coefficient to form a three-dimensional device state vector, to calculate a Mahalanobis distance between the device state vector and a preset working condition reference cluster to obtain an anomaly deviation value of the associated device, and to determine that the corresponding associated test device is a first anomaly-associated device when the anomaly deviation value is greater than a preset first anomaly deviation threshold value.

5. The device automation detection system based on multi-modal perception fusion according to claim 4, characterized in that, The device determination module is configured to construct a three-dimensional time sequence feature matrix based on the vibration intensity back difference, the micro-temperature rise gradient, and the baseline drift rate within the same preset abnormality determination duration, to calculate singular value decomposition of the time sequence feature matrix, and to take the maximum singular value as a system coupling strength, and to determine that the corresponding associated test device is the second abnormal associated device when the system coupling strength is greater than a preset coupling strength threshold.

6. The multi-modal perception fusion based equipment automation detection system according to claim 5, wherein, The adjustment module is configured to construct an abnormal device set including all the first abnormal associated devices and the second abnormal associated devices, to extract direct connection relationships and indirect connection relationships between the associated test devices in the abnormal device set based on the test linkage relationship chain to form an abnormal propagation subgraph, and to calculate betweenness centrality of each node in the abnormal propagation subgraph and a correction index according to the betweenness centrality and the corresponding abnormal strength. The abnormal strength is the abnormal deviation value or the system coupling strength.

7. The device automation detection system based on multi-modal perception fusion of claim 6, wherein, The adjustment module is configured to increase a preset feeding rate in the preset detection parameter according to a relative deviation of the correction index and a maximum value of the preset calibration range when the correction index is greater than the maximum value of the preset calibration range, and to decrease a preset excitation force in the preset detection parameter according to a relative deviation of the correction index and a minimum value of the preset calibration range when the correction index is less than the minimum value of the preset calibration range.

8. The multi-modal perception fusion based equipment automation detection system of claim 7, wherein, The correction module is configured to calculate Jaccard similarity coefficients of all the first abnormal associated devices and the second abnormal associated devices within the preset correction period to obtain a plurality of static coincidence degrees, to calculate a mean component, a trend component, and a fluctuation component of all the static coincidence degrees, and to correct the preset abnormality determination threshold according to the mean component, the trend component, and the fluctuation component. The trend component is calculated by linear regression on all the static coincidence degrees, and a regression coefficient is taken as the trend component.

9. The multi-modal perception fusion based equipment automation detection system according to claim 8, wherein, The correction module is configured to input the mean component, the trend component, and the fluctuation component into a preset fuzzy reasoning system to obtain a comprehensive correction factor, and to correct the preset abnormality determination threshold according to the comprehensive correction factor.

10. A device automatic detection method based on multi-modal perception fusion, applied to the device automatic detection system based on multi-modal perception fusion in any one of claims 1-9, characterized in that, It comprises: Based on a preset neural network, the flow stability of a feeding pipe, the pulsation amplitude of a feeding pump, the vibration intensity back difference, and the micro-pressure difference value of a linkage pressure interface, the consistency coefficient of the particle size distribution of output materials, and the dispersion uniformity of a sample dispersion section in a large health field measurement test based on a preset detection parameter running target test device and each associated test device are obtained in real time. The micro-pressure difference value, the dispersion uniformity, and a preset abnormality determination threshold are used to determine whether a flow path type abnormality or a device performance abnormality occurs. Based on the flow path type abnormality, a plurality of first abnormal associated devices are determined according to the flow stability, the pulsation amplitude, and the consistency coefficient. Based on the device performance abnormality, a plurality of second abnormal associated devices are determined according to the vibration intensity back difference, the micro-temperature rise gradient, and the baseline drift rate. determine a correction index according to the linkage relationship of the first abnormal correlation device and the second abnormal correlation device, and adjust the preset detection parameter according to the correction index; correct the preset abnormality determination threshold according to the coincidence feature of the first abnormal correlation device and the second abnormal correlation device re-determined in a preset correction period after the preset detection parameter is adjusted.

Images