A sensor signal acquisition processing system and method

Abstract

The present application relates to the technical field of signal acquisition, in particular to a sensor signal acquisition processing system and method, comprising: connecting the current sensor and the upper sensor through a link, reading the sensor state sequence of each level sensor in the whole link; according to the sensor state sequence of the previous moment, performing state recognition on the current sensor state sequence, based on the distribution position of each level sensor, converting the state of each sensor into state transition data; performing probability modeling on the state transition data, determining the signal components under each level sensor in the form of entropy screening; defining the fluctuation range of each signal component, and performing merging condition judgment in the fluctuation range of the signal component, performing fault query in the merging result of the signal component, and determining the fault classification of each level signal component. The dynamic response capability and data integrity during signal acquisition processing are realized.

Description

Technical Field

[0001] This invention relates to the field of signal acquisition technology, specifically a sensor signal acquisition and processing system and method. Background Technology

[0002] In the monitoring and control systems of large-scale equipment in bridge and tunnel projects, multiple sensors are typically deployed on the same plane to collect parameters such as position, distance, temperature, and pressure in real time. Traditional solutions require each sensor to be independently connected to a data acquisition unit, leading to the following problems: complex wiring: The large number of sensors necessitates numerous cables to the acquisition unit, resulting in high installation and maintenance costs; high acquisition port occupancy: Each sensor occupies one acquisition channel, causing a shortage of interface resources on the main control equipment; signal attenuation and voltage drop: When sensors are connected in series over long distances, the signal is easily affected by line impedance, resulting in voltage drop and distortion.

[0003] For example, Chinese Patent Publication No. CN114935887A discloses a distributed signal acquisition device and a launch vehicle. The device includes multiple signal acquisition modules. The signal acquisition modules are distributed in each stage of the launch vehicle and connected to sensors in each stage to acquire measurement signals from the sensors in each stage. The signal acquisition module of the current stage is connected to the signal acquisition module of the previous stage via LVDS. It is used to send the measurement signal acquired by the current stage signal acquisition module to the signal acquisition module of the previous stage based on the signal acquisition time, and to send a time synchronization request to the signal acquisition module of the previous stage and receive a time synchronization response from the signal acquisition module of the previous stage. Based on the sending and receiving times of the time synchronization request and the sending and receiving times of the time synchronization response, the signal acquisition time is adjusted.

[0004] For example, Chinese Patent Publication No. CN115657557A discloses a signal acquisition system. The system includes: acquiring a voltage signal through a signal acquisition module, transmitting the acquired voltage signal to a burn detection module, generating a corresponding digital indication signal based on the comparison result between the acquired voltage and a reference voltage, transmitting the digital indication signal to a control module, and determining the corresponding burn detection result based on the digital indication signal. The system utilizes the acquired voltage signal acquired by the signal acquisition module to perform burn detection judgment.

[0005] In existing technologies, the time of signal acquisition is synchronized to ensure that the received data are in the same time dimension. Then, by identifying the analog voltage, the burn-out status of each device under the current signal acquisition is determined. However, existing technologies do not consider the transmission of data acquired by multiple sensors after connection, which can easily lead to abnormalities that are difficult to locate, processing delays, and data misjudgments when multiple sensors are cyclically transmitting data. Summary of the Invention

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a sensor signal acquisition and processing system, comprising: a signal acquisition module, used to acquire signal data sent by the sensor, connect the current sensor and the upper-level sensor through a link, and read the sensor state sequence of each level sensor in the entire link.

[0007] The state transition module is used to identify the current sensor state sequence based on the sensor state sequence of the previous moment, and to assemble the state transitions of each sensor into state transition data based on the distribution location of each level of sensor.

[0008] The signal decomposition module is used to perform probabilistic modeling on state transition data and determine the signal components of each sensor level by using entropy filtering.

[0009] The output analysis module is used to define the fluctuation range of each signal component, and to determine the merging conditions based on the fluctuation range of the signal components, and select the merging strategy for the signal components under each level of sensor.

[0010] The fault classification module is used to obtain the merged result of the signal components based on the merging strategy of the signal components under each level of sensor; and to perform fault query based on the merged result of the signal components to determine the fault classification of each level of signal components.

[0011] A sensor signal acquisition and processing method includes: S1, acquiring signal data sent by the sensor, connecting the current sensor with the upper-level sensor through a link, and reading the sensor state sequence of each level of sensor throughout the entire link.

[0012] S2, based on the sensor state sequence of the previous moment, perform state identification on the current sensor state sequence, and based on the distribution location of each level of sensor, assemble the state transition of each sensor into state transition data.

[0013] S3 determines the signal components of each sensor level by probabilistically modeling the state transition data and using entropy filtering.

[0014] S4 defines the fluctuation range for each signal component and uses the fluctuation range of the signal component to determine the merging condition, and selects the merging strategy for the signal components under each level of sensor.

[0015] S5. Based on the merging strategy of signal components under each level of sensor, obtain the merged result of the merged signal components; use the merged result of the signal components to perform fault query and determine the fault classification of each level of signal components.

[0016] The beneficial effects of this invention are as follows: First, this invention connects the current sensor with the upper-level sensor through a multi-level sensor link to form a chain topology; when the acquisition time is consistent, the signal data is directly combined; when the time is inconsistent, a sensor state sequence is generated by time difference marking and lateral alignment; for multiple sets of upper-level data, an overlay method is used to merge the data to ensure the time consistency of the sensor state sequence and maintain the high integrity of the current sensor state sequence.

[0017] Second, this invention defines the effective data range based on the overlap duration and similarity of signals from the previous and current moments, and generates state transition curves according to the sensor positions. It uses data under different state transition changes as output state transition data to illustrate the dynamic changes of sensor-collected data under multiple data dimensions, emphasizes the transmission of data state changes, and provides a data foundation for subsequent probability modeling and fault detection.

[0018] Third, this invention standardizes the current sensor configuration order and data combination order, and then uses a vector space model and dependency analysis to match similar sequences from the historical database, calculates the distribution entropy value of the state transition data, and filters high-confidence paths. Finally, it determines the optimal path through cross-calculation, generates signal components, and obtains the local optimal solution under signal data transmission to determine the state transition process under the current sensor acquisition data.

[0019] Fourth, this invention sets a fluctuation range for each signal component based on historical confidence intervals and uses merging conditions to sequentially merge and query the signal components, thereby completing the correlation analysis of the merging results; ultimately improving the response rate and processing accuracy for fault handling. Attached Figure Description

[0020] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0021] Figure 1 This is a system framework diagram of a sensor signal acquisition and processing system.

[0022] Figure 2 This is a schematic diagram of a sensor signal acquisition and processing system.

[0023] Figure 3 This is a flowchart illustrating the signal acquisition module of a sensor signal acquisition and processing system.

[0024] Figure 4 This is a flowchart illustrating the state transition module of a sensor signal acquisition and processing system.

[0025] Figure 5 This is a flowchart illustrating the signal decomposition module of a sensor signal acquisition and processing system.

[0026] Figure 6 This is a flowchart illustrating the output analysis module of a sensor signal acquisition and processing system.

[0027] Figure 7 This is a flowchart illustrating the fault classification module of a sensor signal acquisition and processing system.

[0028] Figure 8 This is a flowchart illustrating a sensor signal acquisition and processing method.

[0029] In the diagram: 1. Power supply; 2. Distributed edge computing unit; 3. Anomaly detection unit; 4. Front-end heterogeneous coupler unit; 5. Dynamic impedance matching unit. Detailed Implementation

[0030] The embodiments of the present invention are described in detail below. The embodiments described below are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention. Where specific techniques or conditions are not specified in the embodiments, they shall be performed in accordance with the techniques or conditions described in the literature in the art or in accordance with the product manual.

[0031] See Figure 1 A sensor signal acquisition and processing system includes: a signal acquisition module, a state transition module, a signal decomposition module, an output analysis module, and a fault classification module; wherein, the output terminal of the signal acquisition module is connected to the state transition module, the output terminal of the state transition module is connected to the signal decomposition module, the output terminal of the signal decomposition module is connected to the output analysis module, and the output terminal of the output analysis module is connected to the fault classification module.

[0032] The signal acquisition module is used to acquire signal data sent by the sensors, connect the current sensor with the upper-level sensor through the link, and read the sensor status sequence of each level of sensor throughout the entire link.

[0033] The state transition module is used to identify the current sensor state sequence based on the sensor state sequence of the previous moment, and to assemble the state transitions of each sensor into state transition data based on the distribution location of each level of sensor.

[0034] The signal decomposition module is used to perform probabilistic modeling on state transition data and determine the signal components of each sensor level by using entropy filtering.

[0035] The output analysis module is used to define the fluctuation range of each signal component, and to determine the merging conditions based on the fluctuation range of the signal components, and select the merging strategy for the signal components under each level of sensor.

[0036] The fault classification module is used to obtain the merged result of the signal components based on the merging strategy of the signal components under each level of sensor; and to perform fault query based on the merged result of the signal components to determine the fault classification of each level of signal components.

[0037] When forming the sensor state sequence of the entire link, the sensor state sequence collects data from each sensor and the combination of the upper-level sensors, and outputs the sensor state sequence step by step according to the time order of their combination, so as to illustrate the implementation process of the current sensor in the link-based processing. This step is to obtain the sensor state sequence of multi-level signals by superimposing and transmitting signals.

[0038] It should be noted that the sensors involved in the current scenario include, but are not limited to, laser displacement gauges, ultrasonic sensors, thermocouples, infrared sensors, and piezoresistive sensors. These sensors will represent the collected spatial coordinates, deformation data, temperature field data, and structural stress data, converting this data into signal form and completing long-distance transmission through multi-level sensor conduction.

[0039] like Figure 3 As shown, the implementation of the signal acquisition module also includes: receiving signal data from the previous level sensor, and when the acquisition time of the previous level sensor is consistent with the acquisition time of the current sensor, combining the current signal data with the previous level signal data into a sensor state sequence.

[0040] When the acquisition time of the previous sensor is inconsistent with the acquisition time of the current sensor, the time difference between the previous sensor and the current sensor is received, and the horizontal mark of the time axis verification is retrieved. The signal data after the horizontal mark is output as a sensor state sequence.

[0041] When performing horizontal labeling, the data from the previous sensor and the current sensor are mapped to adjacent time points on a unified time axis. For example, if they are inconsistent, if the time difference is less than 10ms, they will be updated to synchronous time points and the corresponding data will be merged. At the same time, data that is greater than 10ms will be marked as asynchronous time points and these data will be input into the subsequent data analysis section.

[0042] This part of the processing will focus on the distributed edge computing process of each sensor. If the amount of data received is too large, the received data will be preprocessed and divided into multiple data segments according to the order of each sensor combination. These data will then be sent to the external control terminal, which will process the sensor state sequence of these sensor signal data combinations.

[0043] Simultaneously, when receiving signal data, the implementation method also includes: when the currently acquired signal data corresponds to multiple sets of superior signal data, for the received superior signal data, the number of data blocks of the superior signal data is counted. When the number of data blocks is greater than a preset threshold, the superior signal data is superimposed with the current signal data, using the currently acquired signal data as the superposition starting point, and the superimposed data is regarded as the output signal data.

[0044] By dividing the upstream signal data into fixed-size data blocks, such as the number of sampling points within a time window, each data block contains a timestamp, signal value, and metadata (such as sensor ID, acquisition time, etc.). If the data transmitted from multiple sensors to the current data is uniformly sampled, that is, there is no time overlap between the upstream signal data and the current signal data, a weighted superposition method is adopted. A weight is set for each data block, and the weight of each data block is based on the ratio of the signal-to-noise ratio of the corresponding data block to the signal-to-noise ratio of all superimposed data blocks. The weighted superposition of data blocks is completed sequentially, and the weighted superimposed data is regarded as the output signal data. At this time, the weighted superimposed data can reduce the noise in the currently acquired data.

[0045] If the upstream signal data overlaps with the current signal data in time, zero-padding is used to perform segmented convolution and summation. The summed data is then considered the output signal data. The transmitted signal data is divided into N segments of fixed length, each containing N sampling points. Each segment is zero-padding to a length L, where L = N + M - 1, and M represents the filter length. Then, each zero-padding segment is linearly convolved with the zero-padding filter. The convolution result of each segment has a length of L and contains M - 1 overlap points, i.e., the overlap between the end of the current segment and the beginning of the next segment. The convolution operation causes signal expansion, resulting in temporal overlap between the end of the current segment and the beginning of the next segment.

[0046] During the overlapping and addition process, the first N points of each convolution result, excluding the overlapping portion, are directly concatenated to the final output sequence. Then, the last M-1 points corresponding to the overlapping portion of each convolution result are added to the first M-1 points of the next convolution result. This process is repeated until all segments are processed. After the convolution results of all segments are overlapped and added, they are concatenated to form a complete output signal, resulting in the final data superposition.

[0047] Setting a preset threshold for the number of data blocks is essentially about balancing data integrity and processing efficiency. It prevents data transmission instability caused by a large number of sensors during long-distance data transmission. The preset threshold can be set based on the average number of data blocks uploaded by sensors in historical data to balance the disturbances and missing data caused by long-distance transmission.

[0048] When the currently acquired signal data corresponds to only one set, the current signal data is directly combined with the upper-level signal data and directly input to the external control device.

[0049] In one embodiment of the present invention, the state transition module is used to check whether the signal data input by the previous sensor contains abnormal signal data representation, and the abnormal part that exists in the state transition process of the previous sensor when integrating signal data in a link.

[0050] like Figure 4 As shown, the implementation of the state transition module includes: when acquiring the sensor state sequence of the previous moment, defining the effective signal range based on the similarity of the signal data according to the length of the overlapping time period between the upper-level signal data corresponding to the previous moment and the current signal data, extracting multi-dimensional features of each level of sensor according to the effective signal range, and forming a state transition curve according to the arrangement order of the sensors.

[0051] When multiple sensors transmit data in a chain structure, the overlap time between the signal data transmitted from the upper level and the signal data collected by the current sensor is compared to determine data temporal consistency. Then, the Pearson correlation coefficient is calculated for the signal data from the upper and current sensors. If the similarity is greater than 0.8, the data received within the current overlapping time period is considered valid. It should be noted that the signal data collected includes, but is not limited to, numerical values ​​such as temperature and pressure. When extracting multi-dimensional features from each sensor, the moving average of each sensor's output data is used to set corresponding dimensional features, forming state transition curves related to the average values ​​of temperature and pressure. These curves illustrate the relationship between the sensor data collected in the tunnel or other long-distance transmission and the relative position of the sensors. Simultaneously, all relevant data are normalized when calculating the Pearson correlation coefficient.

[0052] The method for calculating the moving average is to calculate the average value of data such as temperature and pressure collected within the overlapping time period, and to distinguish the different average values ​​of data collected from each sensor.

[0053] As for the implementation of multi-dimensional features, it also includes: extracting the data dimensions corresponding to the state transition curves based on the distribution locations represented by sensors at all levels. The data dimensions include, but are not limited to, the acquisition location dimension, displacement dimension, temperature dimension, pressure dimension, spatial location classification dimension, and time dimension.

[0054] The acquisition location dimension describes the spatial coordinates of the sensor data; the displacement dimension describes the displacement and deformation values ​​of the acquired data location; the temperature dimension represents the acquired temperature; and the pressure dimension describes the acquired structural stress data. The spatial location classification dimension describes whether the sensors are deployed on the same plane or in different areas, illustrating the different deployment locations of each sensor. The time dimension emphasizes the timestamps of the sensor's transmitted signal data, illustrating the dynamic process of state transition.

[0055] Determine the minimum data set for each data dimension, and combine the data that satisfy the minimum data set at the current time as the output multi-dimensional features.

[0056] The aforementioned minimum data set represents the minimum amount of data required for analysis in the corresponding dimension. For example, the current sensor obtains temperature, pressure, and displacement signal data through transmission from the upper level and its own acquisition. However, the displacement signal data is insufficient to complete an edge computing operation. In this case, feature extraction is performed on the temperature and pressure components to obtain multi-dimensional features of temperature and pressure. The state transitions of temperature and pressure are then used as output data. The displacement data is then passed to the next level sensor until a certain amount of data is collected and processed. At this point, the minimum data set is determined based on the average amount of data collected in each historical calculation in the corresponding dimension to determine whether the amount of data in the current dimension meets the minimum data set requirement.

[0057] For the acquisition location dimension, the focus is on labeling the total amount of data collected at a single or multiple coordinates, rather than directly calculating it in the edge computing part corresponding to the current sensor; the displacement, temperature, and pressure dimensions will use continuously received data to determine the state transition process at multiple time points; the spatial location classification dimension and the time dimension are used to determine whether the total amount of data collected in the corresponding space and time meets the requirements.

[0058] When acquiring the sensor state sequence at the current moment, the state transition curves at the current moment and the previous moment are compared to determine the rate of change of the state transition curves in consecutive moments. Based on the rate of change of the state transition curves, the state transition data of each level of sensor when receiving data from the previous level sensor is determined.

[0059] The rate of change of the state transition curve at this point is the rate of change of the output data of each sensor at multiple time points to determine the changes at the current acquisition location, so as to promptly detect anomalies in long-distance transmission. The output state transition data will output the data of the state transition curve within the effective signal range and record the rate of change at the preceding and following time points to illustrate the current state change.

[0060] In one embodiment of the present invention, the signal decomposition module further decomposes the state transition data for each sensor to solve the state of the signal data under the link transmission on each sensor, and measures the information transmission process according to the calculation method relative to information entropy.

[0061] like Figure 5 As shown, the implementation of the signal decomposition module includes: matching similar sensor configuration sequences from historical data based on the configuration order and data combination order of the upper-level sensors in the current state transition data.

[0062] Probabilistic modeling is performed on the matched data. Using an information matrix as the carrier, the current state transition data and the sensor configuration sequence are taken as input data. The information entropy value of this distribution is calculated and used as the confidence index for the sensor state transition. By filtering the confidence index, only high-confidence data is retained. Based on the high-confidence data, a path set for the sensor combination is formed to complete subsequent processing.

[0063] Based on the confidence index value, the path set of sensor combinations is filtered. Through the path set of each sensor combination, cross-calculation is performed on the path set to determine the optimal path of the path set during state transition. The optimal path is then used as the signal component corresponding to the current sensor.

[0064] At this point, assuming the transmitted signal data conforms to a specific distribution, such as the Matrix Fisher distribution, the sensor configuration sequence is obtained by matching the current state transition data with historical data. The configuration order of the upper-level sensors includes parameters such as the sensor ID list, data acquisition frequency, and trigger threshold. The data combination order includes signal data under various data combinations such as temperature-pressure joint analysis and position-vibration time series correlation.

[0065] The methods for matching similar sensor configuration sequences from historical data include: standardizing the data corresponding to the configuration order and data combination order of the upper-level sensors, respectively.

[0066] When standardizing the data included in the configuration sequence of the upper-level sensors, the sensor configuration sequence is converted into a vector space model, and the sensor IDs are represented in a data-driven form to determine the data type collected by each sensor. The acquisition frequency and trigger threshold values ​​are normalized; the trigger threshold here can represent the threshold corresponding to the number of data blocks mentioned above. These data are standardized in vector form to form a standardized data format consistent with the current state transition data. Simultaneously, the trigger threshold can also represent the number and distance of sensor configurations in the current state transition data, illustrating the order of sensor configurations within the scenario.

[0067] The vector space model represents mapping the signal data collected by the current sensor, such as ID, data type, acquisition frequency, trigger threshold, etc., into a vector space to form vectorized data.

[0068] When standardizing data corresponding to the data combination order, the data collected by the combined temperature-pressure, position-vibration, and other sensors are standardized according to the dependency relationship between the collected data.

[0069] Then, similarity calculations are performed on the data formats corresponding to the configuration order and data combination order of the upper-level sensors, respectively. That is, the similarity between the current state transition data and historical data is calculated. The similarity is calculated using the cosine similarity method. The average similarity is calculated according to the configuration order and data combination order of the upper-level sensors to identify the current sensor configuration sequence.

[0070] It should be noted that the similarity between the sensor configuration sequence and the current state transition data can be based on a value of 0.6 to select signal data related to the configuration scenario, or the average similarity of the state transition data in historical data used when selecting the sensor configuration sequence can be used as the basis for selecting the sensor configuration sequence in the current scenario.

[0071] When performing probabilistic modeling, the Matrix Fisher distribution is a representation of the Fisher information matrix, and its corresponding probability density function is... As shown below.

[0072] ;in, Represents the normalization constant; The matrix represents the state transition data and sensor configuration sequence input form of the sensor. The matrix is ​​represented in the form of 3×3. At this time, the state transition data and sensor configuration sequence are split into a three-dimensional array and the parameters of the normal distribution under the setting information matrix are extracted. The distribution state of the current signal data transmitted under multiple sensors is described by using the matrix inner product form. The parameters of the Matrix Fisher distribution are represented by a 3×3 matrix. Representation matrix The matrix after matrix transpose; express and Perform inner product; This represents an exponential function with a base of constant e. When creating a probabilistic model, this describes the distribution of the matrix variable R under a given F, indicating the degree of concentration, directional preference, and other characteristics of the distribution.

[0073] Then, based on the obtained probability density function, the information entropy is set to determine the confidence level calculated under the current data. The lower the information entropy value, the more stable the data and the higher the confidence level. The higher the information entropy value, the greater the data fluctuation and the lower the confidence level. Based on this information entropy value, entropy screening is completed, and high-confidence signal components are obtained.

[0074] Information entropy This is represented as: ;in, Let R represent all three-dimensional matrices. At this point, by calculating the information entropy, we can further verify the current data collection situation.

[0075] After obtaining the information entropy, the average value of the information entropy is used as a threshold. Paths with values ​​less than the average value are considered as confidence indicators for screening, and a path set is generated, such as the path set 2→5→7 and 1→3→6. The numbers used here represent the corresponding sensor numbering order during the calculation.

[0076] Preferably, when cross-validating the path set, the validation content includes, but is not limited to, validating the consistency of entropy values ​​and the continuity of state transitions. The weighted sum after validation is used as the current cross-validation score, and the path set with the largest score is selected as the current output signal component to accurately describe the continuous state after multiple data superposition outputs.

[0077] Entropy consistency Represented as: ;in, The entropy value of the i-th path is determined by statistically analyzing the entropy values ​​of the corresponding data in the set of paths. Entropy is usually used to measure the uncertainty or information content of a path. This represents the average entropy value of all paths. This represents the maximum entropy value across all paths. This represents the minimum entropy value of all paths; This represents the total number of paths, with values ​​ranging from 1 to K.

[0078] Continuity of state transitions Represented as: ;in, This represents the number of state transitions, i.e., the state transition data at multiple time points. If the current state transition data exists at T time points, then the value of m ranges from 1 to T-1. Indicates cosine similarity; This represents the state vector composed of state transition data at adjacent time points j+1; This represents the state vector composed of state transition data at adjacent time j+1, and the process of determining its state transition is based on the calculated state transition data.

[0079] When calculating the cross-validation score using weighted averages, a weight of 0.5 can be used to sum the values ​​corresponding to entropy consistency and state transition continuity to obtain the relevant cross-validation score.

[0080] At this point, after determining the optimal state path and relatively high confidence data, the output signal components extract corresponding signal data such as displacement, temperature, and spatial location from these data, and these signal data will be used as the current output signal components.

[0081] In one embodiment of the present invention, the output analysis module is used to identify the fluctuation situation in the fault condition in the signal classification and determine the manifestation form of the merged fault data, so as to determine the influence of each level sensor on the next level sensor through the link transmission.

[0082] like Figure 6 As shown, the output analysis module is implemented by defining a fluctuation range for each signal component based on its confidence interval in historical data, according to the current input signal component. The confidence interval represents the range of values ​​for collected data such as temperature, displacement, and pressure in historical data. For example, if a 95% confidence interval is selected, the confidence interval is set in the form of mean ± 1.96 standard deviations.

[0083] When combining signal components, methods such as maximum ratio combining, equal gain combining, and selective combining can be used. Maximum ratio combining is based on the signal-to-noise ratio (SNR) of each component, where the weight can be based on the ratio of the SNR to the sum of the SNRs of all components. Equal gain combining is performed by aligning the phases and then combining them with equal weights, which means dividing the component into equal parts according to the number of components to be combined to obtain multiple weight values. Selective combining only selects components within a specific range for combining, such as selecting abnormal data outside the fluctuation range for combining, to identify possible fault problems.

[0084] The merging method is determined based on the fluctuation range of the current signal component. At this time, it will be based on the range of the current value relative to the confidence interval. That is, by comparing the value of the current signal component with the fluctuation range, the corresponding merging method is selected according to the comparison, and the merging strategy is selected according to the corresponding merging method.

[0085] The signal components are grouped according to their fluctuation range. The range identifier of each signal component is queried. When the collected signal component is covered by the confidence interval, a normal range identifier is set. If it is not covered by the confidence interval, an abnormal range identifier is set.

[0086] Based on the range identifier of the current signal component, query the database for the merging conditions of each signal component, and select the merging strategy according to the description of the merging conditions.

[0087] It should be noted that the selection of merging strategy should be based on the following three merging conditions: 1. Signal range determination, 2. Signal reliability, and 3. Phase alignment.

[0088] The signal range determination indicates whether the current signal component's value belongs to the normal range or the abnormal range. The normal range uses maximum ratio combining or equal gain combining, while the abnormal range uses selective combining.

[0089] Signal reliability indicates that the current data is highly dependent on the signal-to-noise ratio (SNR), and it is necessary to calculate the SNR of each signal. That is, after determining that the current signal component belongs to the normal range, the maximum ratio is used for merging. Signal reliability is used to determine whether the maximum ratio merging is used in the current scene.

[0090] Phase alignment indicates that signals need to be phase aligned to achieve synchronous sampling; otherwise, equal-gain combining is not performed. Under normal circumstances, the currently acquired signal components will not simultaneously satisfy the combining conditions corresponding to maximum ratio combining and equal-gain combining. When both maximum ratio combining and equal-gain combining are satisfied, equal-gain combining is preferred to avoid deviation from the optimal combining result due to amplification of signal-to-noise ratio errors.

[0091] In one embodiment of the present invention, in the fault classification module, the existence of a fault is determined based on the merging result, and the cost is quantified; the state transition process of the sensor is converted into a minimum fault-based processing method, and the signal components under the optimal state transition path are selected. The signal data is analyzed based on the result of the signal component merging to determine the fault problems existing in the deployment of the multi-level chain sensor.

[0092] like Figure 7 As shown, the implementation of the fault classification module includes: scanning the merging results of signal components, selecting fault labels corresponding to the current signal component from historical data in the form of error statistics, performing cluster analysis on the merging results, converting the data corresponding to the fault labels and the data in the merging results into standard data form during cluster analysis, and setting fault labels for each cluster center. This processing method can quickly quantify the degree of signal anomaly, discover the implicit patterns of signal distribution through unsupervised learning, automatically adapt to environmental changes and sensor characteristic drift, without relying on a large number of manually labeled fault labels, and avoid the failure of sensor threshold settings under fixed paths.

[0093] Preferably, in the above-mentioned cluster analysis, K-means or other clustering methods are used, with the termination condition being that the change in cluster centers in continuous iterations is less than a threshold. That is, after each clustering, the rate of change of the cluster center value compared to the previous clustering is less than a specific threshold. In this case, ±5% will be used as the termination condition to complete the iteration of cluster centers. At the same time, the merged results of the signal components are clustered according to their corresponding temperature, pressure, displacement, and other data during clustering, and the fault labels after clustering indicate the abnormalities existing in the current signal components.

[0094] The system counts the occurrence frequency of each fault tag. When the occurrence frequency of a fault tag reaches a set number, it determines the upstream sensor and the current sensor to which each fault tag belongs. At this time, it identifies the occurrence frequency of the fault tag on the current sensor. Then, when the occurrence frequency of the fault tag reaches a set number, it triggers a combined analysis of the upstream sensor and the current sensor to determine whether the fault is transmitted from the upstream sensor or caused solely by the current sensor itself.

[0095] At this point, the number of occurrences can be set based on the average number of occurrences of faults in historical data. The average number of faults is obtained by extracting the occurrence number of the fault tag from historical data and dividing it by the total time range. This average number of faults represents the number of occurrences within a certain time range. When the current fault tag exceeds the set number of occurrences, it is necessary to analyze the sensors in pairs to determine whether the fault occurred during data transmission or was caused by a single sensor.

[0096] When the upper-level sensor corresponds to at least one fault label, the fault label of the current sensor is correlated with the fault label of the upper-level sensor, and the fault labels of the upper-level sensor and the current sensor after correlation analysis are used as the output fault classification.

[0097] It should be noted that the correlation analysis analyzes the fault labels to determine whether the current sensor is affected by the upper-level sensor. For example, if the current sensor shows an abnormal temperature and the upper-level sensor shows an abnormal data, the current sensor may be faulty after the data is merged due to the fault of the upper-level sensor.

[0098] At this point, the analysis can be performed by predefining a series of rules in the database based on domain knowledge. These rules will adopt the IF-THEN form and judge based on the fault labels of the current sensor and the fault labels of the upper-level sensor to explain whether the fault labels of the upper-level sensor cause abnormalities in the output data of the current sensor after the upper-level sensor sends the input to the current sensor. The output will be grouped according to the direct and indirect association between the upper-level sensor and the current sensor under the fault labels to explain the faults that exist after the current sensor signal is acquired.

[0099] A simple example can be used to illustrate the association analysis rules set. Assuming sensor A is the parent sensor and sensor B is the current sensor, if sensor A has "data anomaly" AND sensor B has "temperature anomaly" within the time window Δt, then the association type is "data transmission failure". At this point, we can know the fault status of the current sensor after it has collected data. Afterwards, we can also quickly locate the sensor's position based on the signals emitted by the sensor itself.

[0100] When the fault label of the upper-level sensor does not correspond, the fault label of the current sensor will be used as the output fault category.

[0101] In one embodiment of the present invention, such as Figure 2 As shown, the sensor signal acquisition and processing system consists of a power supply 1, a distributed edge computing unit 2, an anomaly detection unit 3, a front-end heterogeneous coupler unit 4, and a dynamic impedance matching unit 5. These five devices serve as the sensor data acquisition units, connecting multiple acquisition devices to achieve chain-like data acquisition and transmission across multiple sensor levels.

[0102] During signal acquisition, the output signal of the previous level sensor combination and the signal data of the current sensor are acquired through the front-end heterogeneous coupler unit 4. The sensors are connected in a multi-level chain structure to achieve long-distance signal transmission and acquisition in series. The acquired data supports various types, including but not limited to analog voltage signals, analog current signals, and digital signals.

[0103] Power supply 1 is used to provide sufficient power to the sensor to prevent the sensor from being unable to output a signal due to excessive voltage drop caused by being too far from the control power supply.

[0104] The distributed edge computing unit 2 is used to process the output signal of the previous level sensor and the currently acquired sensor signal; when it reaches the last level, it directly outputs the signal to the PLC controller or other controllers to determine the execution status of the device sensors.

[0105] The anomaly detection unit 3 is used to identify signal problems in the sensor transmission signals. Since it uses a chain to collect signals from multiple sensors, it only outputs one result signal to the controller. It only outputs a presence signal when all sensor signals are present. If any signal is faulty, it will not output a signal to the controller. Therefore, this anomaly detection unit 3 is used in scenarios where there are faults in multi-level sensors. Through this unit, the fault location of the sensor can be accurately located, which is convenient for maintenance and replacement.

[0106] The dynamic impedance matching unit 5 is used to perform dynamic impedance matching on the collected data to eliminate signal distortion in cascaded transmission. The distortion types include distortion, voltage drop, white noise, etc., so that the signal to be processed in the subsequent calculation remains in a standard format. At this time, the signal processing method is the existing technology, which uses ADC / DAC, filters, gain amplifiers, etc. to process the signal data received by multi-level sensors and reduce noise during long-distance signal transmission.

[0107] At this time, the sensor uses the signal output interface on the front-end heterogeneous coupler unit 4 to receive the sensor signal from the previous stage and the current sensor signal, and receives the abnormal detection signal from the previous stage according to the abnormal detection unit 3.

[0108] Since the sensor detects the same state of the device, under normal circumstances, the sensor detection signal is either all high level or all low level. When an inconsistency is found, it means that the sensor has a problem.

[0109] When all connected sensors detect a high output level, the signal output by the first-stage sensor to the control cabinet is high; when at least one connected sensor fails to detect, the signal output by the first-stage sensor to the control cabinet is low.

[0110] When a low output level occurs, the control cabinet can determine which sensor number is faulty by reading the analog voltage and dividing it by a 0.25V base voltage. Based on the multiple relationship between the voltage and the base voltage, it can be determined that a maximum of 100 sensors can be connected in a chain.

[0111] These units comprise the anomaly detection, edge distribution computation, data acquisition, and signal standard processing components of the current system implementation, illustrating the equipment foundation upon which the current system operates.

[0112] like Figure 8 As shown, the present invention also provides a sensor signal acquisition and processing method, including: S1, acquiring signal data sent by the sensor, connecting the current sensor with the upper-level sensor through a link, and reading the sensor state sequence of each level sensor in the entire link.

[0113] S2, based on the sensor state sequence of the previous moment, perform state identification on the current sensor state sequence, and based on the distribution location of each level of sensor, assemble the state transition of each sensor into state transition data.

[0114] S3 determines the signal components of each sensor level by probabilistically modeling the state transition data and using entropy filtering.

[0115] S4 defines the fluctuation range for each signal component and uses the fluctuation range of the signal component to determine the merging condition, and selects the merging strategy for the signal components under each level of sensor.

[0116] S5. Based on the merging strategy of signal components under each level of sensor, obtain the merged result of the merged signal components; use the merged result of the signal components to perform fault query and determine the fault classification of each level of signal components.

[0117] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention, which are still covered within the protection scope of the present invention.

Claims

1. A sensor signal acquisition and processing system, characterized by, The method comprises the following steps: A signal acquisition module is used to acquire signal data sent by a sensor, connect the current sensor with a superior sensor through a link, and read the sensor state sequence of each level sensor in the entire link; A state transition module is used to identify the current sensor state sequence according to the sensor state sequence at the previous moment, and transfer the state of each sensor to form state transition data based on the distribution position of each level sensor; A signal decomposition module is used to probabilistically model the state transition data, determine the signal components under each level sensor in the form of entropy screening, and combine the signal components under each level sensor according to the fluctuation range of each signal component to determine the merging strategy of the signal components under each level sensor; An output analysis module is used to define the fluctuation range of each signal component, and determine the merging strategy of the signal components under each level sensor according to the fluctuation range of the signal components; A fault classification module is used to obtain the merging result of the signal components according to the merging strategy of the signal components under each level sensor, and determine the fault classification of each level signal component according to the merging result of the signal components. The implementation of the signal acquisition module further comprises the following steps:

2. The sensor signal acquisition and processing system of claim 1, wherein, Receiving the signal data of the previous level sensor, and combining the current signal data and the signal data of the previous level sensor into a sensor state sequence when the acquisition time of the previous level sensor is consistent with the acquisition time of the current sensor; When the acquisition time of the previous level sensor is inconsistent with the acquisition time of the current sensor, receiving the time difference between the previous level sensor and the current sensor, and calling the horizontal marker of the time axis verification, and outputting the signal data after the horizontal marker as the sensor state sequence. When receiving the signal data, the implementation further comprises the following steps:

3. The sensor signal acquisition and processing system of claim 2, wherein, When the current acquired signal data corresponds to multiple sets of superior signal data, counting the number of data blocks of the superior signal data for the received superior signal data, and when the number of data blocks is greater than a preset threshold, taking the current acquired signal data as the superposition starting point, superimposing the superior signal data and the current signal data, and regarding the superimposed data as the output signal data; When there is no time overlap between the superior signal data and the current signal data, using weighted superposition to set the weight of each data block, and regarding the weighted superimposed data as the output signal data; If there is time overlap between the superior signal data and the current signal data, using zero padding and segment convolution to add, and regarding the added data as the output signal data. The implementation of the state transition module comprises the following steps:

4. The sensor signal acquisition and processing system of claim 1, wherein, When obtaining the sensor state sequence at the previous moment, defining the effective signal range according to the similarity of the signal data based on the length of the overlapping time period of the corresponding superior signal data and the current signal data at the previous moment, extracting the multi-dimensional features of each level sensor according to the effective signal range, and forming a state transition curve according to the arrangement order of the sensors; When obtaining the sensor state sequence at the current moment, comparing the state transition curves at the current moment and the previous moment to determine the change rate of the state transition curve at consecutive moments, and determining the state transition data of each level sensor receiving the signal data of the previous level sensor based on the change rate of the state transition curve. The implementation of the multi-dimensional features further comprises the following steps:

5. The sensor signal acquisition and processing system of claim 4, wherein, ​ The distribution position of each level sensor is characterized, and a data dimension corresponding to a state transition curve is extracted, including but not limited to a collection position dimension, a displacement dimension, a temperature dimension, a pressure dimension, a spatial position classification dimension and a time dimension. A minimum data set under each data dimension is determined, and data satisfying the minimum data set at the current moment is combined as the output multi-dimensional feature.

6. The sensor signal acquisition and processing system of claim 1, wherein, The implementation mode of the signal decomposition module includes: According to the configuration sequence and data combination sequence of the upper sensor in the current state transition data, a similar sensor configuration sequence is matched from the historical data; The matched data is subjected to probability modeling, and the current state transition data and the sensor configuration sequence are taken as input data in the form of an information matrix, the entropy value of the distribution is calculated as the confidence index of the sensor state transition; According to the value of the confidence index, the path set of the sensor combination is screened, the path set is subjected to cross calculation through the path set of each sensor combination, the optimal path of the path set is determined when the state transition, and the optimal path is taken as the signal component corresponding to the current sensor.

7. The sensor signal acquisition and processing system of claim 6, wherein, The implementation mode of matching a similar sensor configuration sequence from the historical data includes: The data corresponding to the configuration sequence and the data combination sequence of the upper sensor is respectively subjected to standardization processing; When the data included in the configuration sequence of the upper sensor is subjected to standardization processing, the sensor configuration sequence is converted into a vector space model; When the data corresponding to the data combination sequence is subjected to standardization processing, the standardization processing is performed according to the dependency relationship between the collected data; The data forms corresponding to the configuration sequence and the data combination sequence of the upper sensor are subjected to similarity calculation in sequence, the average value of the similarity is calculated according to the configuration sequence and the data combination sequence of the upper sensor, and the current sensor configuration sequence is identified.

8. The sensor signal acquisition and processing system of claim 1, wherein, The implementation mode of the output analysis module includes: According to the current input signal component, a fluctuation range is defined for each signal component in the confidence interval of the signal component in the historical data; Grouping is performed according to the fluctuation range of the signal component, the range identifier of each signal component is queried, a normal range identifier is set when the collected signal component is covered by the confidence interval, and an abnormal range identifier is set if it is not covered by the confidence interval; Based on the range identifier of the current signal component, the merging conditions of each signal component are queried from the database, and a merging strategy is selected according to the description mode of the merging condition.

9. The sensor signal acquisition and processing system of claim 1, wherein, The implementation mode of the fault classification module includes: The merging result of the signal component is scanned, the merging result is selected in the form of error statistics from the historical data corresponding to the current signal component, the merging result is subjected to cluster analysis, and the fault label of each cluster center is set; The occurrence frequency of each fault label is counted, and when the occurrence frequency of the fault label reaches a set number of times, the upper sensor and the current sensor to which each fault label belongs are determined; When the upper sensor corresponds to at least one fault label, the fault label of the current sensor is associated with the fault label of the upper sensor, and the fault label corresponding to the associated upper sensor and current sensor is taken as the output fault classification. When the upper sensor does not correspond to the fault label, the fault label of the current sensor is taken as the output fault classification.

10. A sensor signal acquisition processing method, characterized by, Comprise: S1, collecting signal data sent by the sensor, connecting the current sensor with the upper sensor through the link, reading the sensor state sequence of each level sensor in the whole link; S2, according to the sensor state sequence of the previous moment, state recognition is carried out on the current sensor state sequence, and based on the distributed position of each level sensor, the state transition of each sensor is composed into state transition data; S3, the signal components under each level sensor are determined in the form of entropy screening by probability modeling on the state transition data; S4, the fluctuation range of each signal component is defined, and the merging condition is judged by the fluctuation range of the signal component, and the merging strategy of the signal component under each level sensor is selected; S5, according to the merging strategy of the signal component under each level sensor, the merging result of the signal component after merging is obtained; The merging result of the signal component is used for fault query to determine the fault classification of each level signal component.

Images