A data fusion and anomaly detection method for a multi-channel precision inertial sensor

Abstract

The application relates to the technical field of precision sensing and detection, and discloses a data fusion and abnormality detection method of a multi-channel precision inertial sensor. The method comprises the following steps: counting all data acquisition channels connected with the sensor, acquiring time-dimension acquisition frequency, transmission delay and space-dimension installation position and other space-time information of each channel; combining preset sensor node configuration information to determine a target receiving data acquisition channel cluster corresponding to each sensor node; pre-processing and configuring the space-time information sent by the cluster, calibrating time deviation, unifying data formats, and counting the space-time information of each node cluster; synchronously calling the operation data analysis of each detection port through an abnormality detection database to be detected port, and performing fusion and abnormality detection management on the space-time information of the cluster. The method can optimize the space-time characteristics of data, reduce data redundancy, enhance abnormality recognition capability, improve system adaptability, and is suitable for high-precision and high-reliability application scenarios.

Description

Technical Field

[0001] This invention relates to the field of precision sensing and detection technology, specifically to a data fusion and anomaly detection method for a multi-channel precision inertial sensor. Background Technology

[0002] In modern industrial measurement, aerospace navigation, and intelligent equipment control, multi-channel precision inertial sensors, with their high-precision sensing capabilities of motion and attitude information, have become crucial devices for achieving precise system operation. As application scenarios increasingly demand higher measurement accuracy and reliability, the application scale of multi-channel precision inertial sensors continues to expand, leading to a corresponding increase in the number of data acquisition channels connected to them, forming complex multi-channel data acquisition networks.

[0003] Currently, in the practical application of multi-channel precision inertial sensors, each data acquisition channel often operates independently, lacking unified statistical analysis and integration of channel spatiotemporal information. Due to differences in the installation location and data transmission rate of different data acquisition channels, deviations occur in the temporal synchronization and spatial correlation of the inertial data acquired by each channel. If these deviations are not addressed, they will directly affect the accuracy of subsequent data fusion, thereby reducing the measurement accuracy of the entire sensing system.

[0004] Current technologies for anomaly detection in multi-channel precision inertial sensor data often employ a single-channel independent detection approach. This involves setting a separate detection threshold for each data acquisition channel to determine if anomalies exist in that channel's data. This approach has significant limitations: it fails to fully utilize the correlation between data from different channels; when multiple channels simultaneously exhibit minor anomalies, single-channel detection struggles to identify such coordinated anomalies, easily leading to missed detections; and due to the lack of a unified anomaly detection management mechanism, the detection results for each channel must be processed separately, increasing data processing complexity and potentially causing false detections due to inconsistent detection standards across different channels, thus affecting the system's efficiency in judging abnormal states.

[0005] Existing technologies lack a dynamic configuration mechanism for the correspondence between sensor nodes and data acquisition channels. When the hardware structure of the sensing system is adjusted, such as adding data acquisition channels or replacing sensor nodes, the correspondence between nodes and channels needs to be manually reconfigured. This operation is cumbersome, error-prone, and reduces the system's flexibility and scalability. These problems make it difficult for existing data processing methods for multi-channel precision inertial sensors to meet the requirements of high-precision and high-reliability applications, thus hindering their further promotion and application in high-end manufacturing, aerospace, and other fields. Summary of the Invention

[0006] The purpose of this invention is to provide a data fusion and anomaly detection method for multi-channel precision inertial sensors to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides a data fusion and anomaly detection method for a multi-channel precision inertial sensor, the method comprising: The data acquisition channels connected to the multi-channel precision inertial sensor are statistically analyzed, and the spatiotemporal information of each data acquisition channel is obtained. Based on the spatiotemporal information of each data acquisition channel and the preset sensor node configuration information, the target receiving data acquisition channel cluster of each sensor node is obtained. The spatiotemporal information transmitted by the target receiving data acquisition channel cluster of each sensor node is preprocessed and configured, and the spatiotemporal information of the cluster of each sensor node is statistically analyzed after the preprocessing configuration. The system retrieves and analyzes the running data of each detection port synchronously from the port to be inspected in the anomaly detection database, and integrates the spatiotemporal information of the cluster for anomaly detection management.

[0008] Preferably, the spatiotemporal information of each data acquisition channel includes the data sampling frequency, data volume, and data transmission rate of each data acquisition channel; The preset sensor node configuration information includes the memory utilization, bandwidth utilization, data processing rate, and CPU utilization of each sensor node.

[0009] Preferably, the processing based on the spatiotemporal information of each data acquisition channel and the preset sensor node configuration information is as follows: Based on the spatiotemporal information of each data acquisition channel, the basic feature values ​​of each data acquisition channel are obtained. The basic feature values ​​of each data acquisition channel are used to quantify the data acquisition efficiency of each data acquisition channel. The data acquisition channels are classified according to sensor type to obtain the data acquisition channels corresponding to each sensor type. The average value of the basic feature values ​​of the data acquisition channels corresponding to each sensor type is taken to obtain the comprehensive feature value of the data acquisition channels corresponding to each sensor type. The data processing energy efficiency characterization value of each sensor node is obtained by comprehensively analyzing the preset sensor node configuration information. The data processing energy efficiency characterization value of each sensor node is used to quantify the data processing capability of each sensor node.

[0010] Preferably, the specific process of obtaining the target receiving data acquisition channel cluster of each sensor node is as follows: The data processing energy efficiency characterization value of each sensor node is matched with the target reception comprehensive characteristic value range of each data acquisition channel corresponding to the data processing energy efficiency characterization value range stored in the information management database. The target reception comprehensive characteristic value range of each data acquisition channel corresponding to the data processing energy efficiency characterization value of each sensor node is statistically calculated and recorded as the target reception comprehensive characteristic value range of each sensor node's data acquisition channel. The comprehensive characteristic value of the data acquisition channel corresponding to each sensor type is matched with the range of the comprehensive characteristic value of the target reception of the data acquisition channel of each sensor node. If the comprehensive characteristic value of the data acquisition channel corresponding to a certain sensor type is within the range of the comprehensive characteristic value of the target reception of the data acquisition channel of a certain sensor node, then the sensor type is defined as the target reception sensor type of the sensor node. The process is repeated to iterate and count all target reception sensor types of each sensor node and all data acquisition channels corresponding to all target reception sensor types. The data acquisition channels corresponding to all target receiving sensor types of each sensor node are uniformly recorded as the target receiving data acquisition channels of each sensor node, thereby integrating the target receiving data acquisition channel cluster of each sensor node.

[0011] Preferably, the preprocessing configuration of the spatiotemporal information transmitted by the target receiving data acquisition channel cluster of each sensor node is as follows: The average value of the basic characteristic values ​​of each target receiving data acquisition channel corresponding to each sensor node is taken to obtain the data transmission evaluation value of each sensor node. The data transmission evaluation value of each sensor node is compared with the set data transmission evaluation threshold. If the data transmission evaluation value of the sensor node is lower than the set data transmission evaluation threshold, the data of the sensor node will continue to be preprocessed and configured with the current data preprocessing parameters. If the data transmission evaluation value of the sensor node is higher than or equal to the set data transmission evaluation threshold, the data of the sensor node will be preprocessed and configured with the preset data preprocessing parameters in the information management database.

[0012] Preferably, the process of statistically analyzing the cluster spatiotemporal information of each sensor node after preprocessing configuration is as follows: After preprocessing and configuration, a data fusion feasibility signal is output. The anomaly detection database receives the data fusion feasibility signal and transmits the cluster spatiotemporal information of each sensor node to the port to be inspected in the anomaly detection database.

[0013] Preferably, the anomaly detection database synchronously retrieves and analyzes the running data of each detection port from the port to be inspected. The specific process is as follows: The operating data of each detection port includes the memory utilization, transmission latency, data read / write speed, and data transmission rate of each detection port. Based on the operating data of each detection port, the detection performance benchmark value of each detection port is obtained. The detection performance benchmark value of each detection port is used to quantify the utilization of the detection capability of each detection port. Collect cluster spatiotemporal information data of each sensor node, including the data processing energy efficiency characterization value of each sensor node and the number of target receiving data acquisition channels. The cluster spatiotemporal information data of each sensor node are comprehensively analyzed to obtain the cluster spatiotemporal information evaluation value of each sensor node. The cluster spatiotemporal information evaluation value of each sensor node is used to quantify the comprehensive performance of each sensor node. The detection performance verification value of each sensor node is obtained by matching the cluster spatiotemporal information evaluation value of each sensor node; The detection performance benchmark value of each detection port is compared with the detection performance verification value of each sensor node. If the detection performance benchmark value of a certain detection port is higher than and closest to the detection performance verification value of a certain sensor node, then the detection port is recorded as the target detection port of that sensor node. The target detection ports of each sensor node are obtained by traversing in turn, and the cluster spatiotemporal information of each sensor node is transmitted to the corresponding target detection port.

[0014] Preferably, the specific process of fusing and managing cluster spatiotemporal information and anomaly detection is as follows: Standard spatiotemporal information curves of multi-channel precision inertial sensors are obtained based on historical databases; Real-time cluster spatiotemporal information curves are obtained through the target detection port; The spatiotemporal information deviation curve is obtained based on the real-time cluster spatiotemporal information curve and the standard spatiotemporal information curve. Determine whether the spatiotemporal information deviation curve matches the preset abnormal pattern characteristics. If they match, generate an anomaly detection result based on the spatiotemporal information deviation curve. If they do not match, continue to monitor the spatiotemporal information deviation curve and perform data fusion management based on the standard spatiotemporal information curve.

[0015] Preferably, the specific process of generating anomaly detection results based on the spatiotemporal information deviation curve is as follows: Analyze the matching degree between the spatiotemporal information deviation curve and the preset anomaly pattern curve to determine the anomaly pattern type; Based on the abnormal mode type and the standard spatiotemporal information curve, the spatiotemporal information prediction curve of the multi-channel precision inertial sensor is obtained. Adjusting data fusion parameters based on spatiotemporal information prediction curves to manage anomaly responses.

[0016] Preferably, the matching degree between the spatiotemporal information deviation curve and the preset anomaly pattern curve is determined by the following process: The error value between the spatiotemporal information deviation curve and the curves of each preset anomaly mode is calculated within the time window. Obtain the preset anomaly pattern curve corresponding to the minimum error value, and denote it as the matching anomaly pattern curve; The minimum error value is converted into a matching degree value, and the matching degree value is inversely correlated with the error value. An anomaly detection report is generated based on the matching degree value and the anomaly pattern curve.

[0017] Compared with the prior art, the beneficial effects of the present invention are: By statistically analyzing the spatiotemporal information of each data acquisition channel, a comprehensive understanding can be achieved of key information such as data acquisition frequency and transmission delay in the time dimension, and installation location and coverage area in the spatial dimension. This provides comprehensive and accurate foundational data for subsequent data processing. Based on this spatiotemporal information and pre-defined sensor node configuration information, a target data acquisition channel cluster for each sensor node is obtained. This enables a reasonable match between sensor nodes and data acquisition channels, ensuring that each sensor node only receives data from channels appropriate to its functional requirements and measurement range. This avoids interference from irrelevant data in the node's data processing, making data processing more targeted, reducing unnecessary data redundancy, and improving data processing efficiency. In the preprocessing configuration stage, the spatiotemporal information transmitted by the target receiving data acquisition channel cluster of each sensor node is preprocessed and configured. This calibrates the time deviation of data from different channels, achieving synchronization of data across channels in the time dimension. Simultaneously, it standardizes the data format and coordinate system in the spatial dimension, eliminating data format incompatibility issues caused by channel differences. The cluster spatiotemporal information obtained after preprocessing is optimized in terms of data format, time synchronization, and spatial correlation, providing a high-quality data foundation for subsequent data fusion. This enables the fusion process to more efficiently uncover the inherent correlations between data from different channels, fully utilize the complementarity of multi-channel data, thereby improving the accuracy and completeness of the fused data and more realistically reflecting the motion state or attitude information of the measured object. By synchronously retrieving operational data from each detection port in the anomaly detection database for analysis, and integrating cluster spatiotemporal information into anomaly detection management, the traditional single-channel independent detection mode has been changed. Synchronously retrieving operational data from each detection port enables collaborative analysis of multi-channel data, fully utilizing the correlation between data from different channels. This not only allows for the rapid identification of obvious anomalies in single-channel data but also accurately captures subtle anomalies occurring collaboratively across multiple channels, effectively preventing missed anomaly detections. Simultaneously, management based on a unified anomaly detection database provides consistent detection standards and analysis frameworks for each detection port, reducing false positives caused by inconsistent detection standards and improving the accuracy of anomaly detection. Furthermore, integrating cluster spatiotemporal information into the anomaly detection process allows for comprehensive judgment based on the spatiotemporal characteristics of the data, further enhancing the ability to identify anomaly data and making the anomaly detection results more reliable. This helps the system promptly detect abnormal states during the operation of the sensing system, enabling staff to take appropriate countermeasures. This method matches sensor nodes with data acquisition channel clusters based on preset sensor node configuration information, offering flexibility and scalability. When the hardware structure of the sensing system is adjusted, simply updating the preset sensor node configuration information automatically regenerates the target receiving data acquisition channel cluster for each sensor node. This eliminates the need for manual adjustment of the node-channel correspondence, simplifying the system adjustment process, reducing the probability of human error, and enhancing the system's adaptability to hardware structure changes. This allows the system to better meet the needs of different application scenarios, expanding the method's applicability. This method, through clustered management of data acquisition channels and preprocessing configuration of data, can reduce the data processing volume of sensor nodes, lower the computational load of nodes, extend the lifespan of sensor nodes, and reduce system energy consumption. During data fusion, fusion is performed based on optimized cluster spatiotemporal information, which reduces computational complexity, improves fusion efficiency, and shortens the data processing cycle, enabling the sensing system to quickly output measurement results and meet real-time application requirements. Attached Figure Description

[0018] Figure 1 This is a schematic diagram illustrating the working principle of the data fusion and anomaly detection method for multi-channel precision inertial sensors described in this invention. Figure 2 A flowchart defined for information; Figure 3 A flowchart for matching channel clusters; Figure 4 Flowchart for preprocessing configuration; Figure 5 This is a flowchart for fusion and anomaly detection. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Please see Figure 1 This invention provides a data fusion and anomaly detection method for multi-channel precision inertial sensors, the method comprising: All data acquisition channels connected to a multi-channel precision inertial sensor are statistically analyzed, and the spatiotemporal information of each channel is acquired. This spatiotemporal information, along with pre-defined sensor node configuration information, is processed to determine the target receiving data acquisition channel cluster corresponding to each sensor node. Next, the spatiotemporal information sent from the target receiving data acquisition channel cluster to each sensor node is preprocessed and configured. After preprocessing, the cluster spatiotemporal information of each sensor node is statistically analyzed. The operational data of each detection port is synchronously retrieved from the port to be inspected in the anomaly detection database for analysis, and the statistically obtained cluster spatiotemporal information is fused and processed for anomaly detection management.

[0021] Example 1: See Figure 2 Consider a multi-channel precision inertial measurement unit (IMU) system containing multiple different types of sensors, such as a three-axis accelerometer, a three-axis gyroscope, and a three-axis magnetometer, collectively forming a nine-axis IMU. Each physical sensor may acquire data through multiple channels to provide redundant or varying levels of precision. For example, a high-precision fiber optic gyroscope might simultaneously provide raw angular velocity data and pre-compensated temperature data, outputting them through two separate channels. The system deploys several sensor nodes, which may be embedded processors, FPGAs, or microcontroller units, responsible for receiving, processing, and performing preliminary analysis of the information from these data acquisition channels.

[0022] For the statistical analysis and information acquisition of all these data acquisition channels, the system first identifies all active data acquisition channels connected to precision inertial sensors. Each channel is assigned a unique identifier and its key spatiotemporal information is recorded. This spatiotemporal information includes the data sampling frequency, i.e., the number of data points generated per second by the channel; the data volume, typically referring to the size of a single data packet or the volume of data generated per unit time; and the data transmission rate, i.e., the actual rate at which the channel transmits data to the node, which may be affected by physical link bandwidth and protocol overhead. For example, a channel used to acquire raw gyroscope data may have a high sampling frequency and a high data volume, but its data transmission rate may be limited due to the use of a low-power wireless protocol. Conversely, a channel transmitting data that has undergone preliminary aggregation and compression may have a smaller data volume, but its transmission rate may be more stable.

[0023] The system accesses pre-defined sensor node configuration information. This information is not real-time measurement values, but rather parameters pre-set based on the hardware specifications and performance indicators of each node, or typical performance values ​​obtained from long-term operational statistics. These include memory utilization for each sensor node, referring to the typical proportion of memory allocated to data processing relative to its total available memory; bandwidth utilization, referring to the typical level of bandwidth usage when the node's data interface processes data streams; data processing rate, referring to the typical speed at which the node's CPU or dedicated processing core processes specific algorithms (such as filtering and computation); and CPU utilization, referring to the typical computational resource usage of the node's main processing unit when running a given task load. For example, a high-performance sensor node might be configured with a high data processing rate and low CPU utilization margin, indicating strong processing capabilities and a moderate current load. Conversely, a resource-constrained node might be configured with high memory utilization and near-saturated bandwidth, indicating more strained resources.

[0024] After acquiring the aforementioned basic information, the system enters the processing phase, aiming to obtain a series of quantified feature values. First, for each independent data acquisition channel, the system calculates a basic feature value based on its spatiotemporal information. This calculation process is not a simple arithmetic addition, but a comprehensive weighted evaluation aimed at quantifying the data acquisition efficiency of that channel. High sampling frequency, large data volume, and high transmission rate usually mean high data throughput, but may also place greater pressure on the receiving node. Therefore, the basic feature value is a scalar value that combines these three parameters according to their importance weights; its level directly reflects the "rate and volume" characteristics of the channel's output data. The system calculates the basic feature value for each data acquisition channel using a weighted linear model. This model comprehensively considers three key spatiotemporal parameters: data sampling frequency, data volume per unit time, and data transmission rate. The specific calculation formula is as follows: ; in, Representing the The basic characteristic values ​​of each data acquisition channel. This represents the data sampling frequency of the channel, after normalization. This represents the average amount of data in the channel per unit time, after normalization. This represents the actual data transmission rate of the channel, after normalization. , , These are the weight coefficients for the corresponding parameters, calibrated using the Analytic Hierarchy Process (AHP). After initialization, they can be adjusted online according to the scenario. All weight coefficients range from 0 to 1 and satisfy the following conditions: The weighting coefficients are pre-set and stored in the information management database during system initialization, based on the relative importance of each parameter's impact on channel data acquisition efficiency through expert evaluation. This calculation formula transforms the spatiotemporal information of the channel into a single, comparable scalar value. This enables the quantification of channel data acquisition efficiency.

[0025] A high fundamental characteristic value for a channel indicates that it is a source with strong data output and transmission capabilities. The system categorizes all data acquisition channels according to sensor type. For example, all channels acquiring acceleration data are categorized as accelerometers, all channels acquiring angular velocity data are categorized as gyroscopes, and so on. For all channels categorized under the same sensor type, the system calculates the arithmetic mean of their individual fundamental characteristic values. This average value is the comprehensive characteristic value of the data acquisition channels corresponding to that sensor type. This value represents the overall average efficiency level of the data acquisition channels for a certain type of sensor (such as all gyroscopes). If a sensor type contains multiple high-performance channels, its comprehensive characteristic value will be higher; conversely, if its channels are mostly configured with low data rates, the comprehensive characteristic value will be lower.

[0026] The system comprehensively analyzes the preset sensor node configuration information to calculate the data processing energy efficiency characterization value for each sensor node. This analysis is also a multi-parameter fusion calculation aimed at quantifying the data processing capability of each node. Node memory utilization, bandwidth utilization, data processing rate, and CPU utilization are all considered. An ideal high-energy-efficiency node should have a high data processing rate while maintaining low memory, bandwidth, and CPU utilization (indicating sufficient resources). Conversely, a node with a moderate processing rate but high resource utilization may have a low energy efficiency characterization value. Therefore, the data processing energy efficiency characterization value is an indicator that comprehensively reflects a node's computing power, resource sufficiency, and processing efficiency. The system calculates the data processing energy efficiency characterization value for each sensor node using a multi-parameter fusion model. Energy efficiency characterization value intervals are divided according to the sensor node's hardware performance level, and comprehensive characteristic value ranges are divided according to the channel data throughput level. The intervals and ranges are continuous and non-overlapping numerical intervals. If no matching node is found for the sensor type's comprehensive characteristic value, the node is automatically assigned to the node with the highest energy efficiency characterization value; if no matching sensor type is found for a node, it is marked as an idle node and enters standby mode.

[0027] This model aims to quantify the comprehensive data processing capability of a node under resource constraints. Its calculation combines information on the node's memory, bandwidth, computing speed, and utilization. The specific calculation formula is as follows: ; in, Representing the Energy efficiency characterization value of data processing for each sensor node This represents the memory utilization rate of the node (value range: 0-1). This represents the bandwidth utilization of the node (value range 0-1). This represents the data processing rate of the node, after normalization. This represents the CPU utilization of the node (value range 0-1). , , These are weighting coefficients used to balance the impact of different resource dimensions on energy efficiency, and they satisfy... The weighting coefficients are pre-set through experimental calibration during system deployment, based on the characteristics of the node hardware architecture and the type of tasks being processed. The formula... The item represents the effective processing capacity with available memory. The term represents the effective processing capacity within the available bandwidth. It directly reflects the activity level of the computing unit. The higher the value, the stronger the overall data processing performance of the node, provided that resources are relatively abundant.

[0028] Example 2: See Figure 3Following the processing results of Example 1, the system now possesses two sets of key data: one set is the comprehensive characteristic value of the data acquisition channels corresponding to each sensor type (such as accelerometer, gyroscope, magnetometer), which represents the average intensity level of the data output from all channels of this type of sensor; the other set is the data processing energy efficiency characterization value of each sensor node, which quantifies the comprehensive data processing capability of each node. The information management database pre-stores the mapping relationship between the data processing energy efficiency characterization value range and the target reception comprehensive characteristic value range of the data acquisition channel. This mapping relationship is predefined based on system architecture design, historical performance data, and resource optimization principles. For example, a node range with extremely high energy efficiency characterization values ​​may map to a very wide comprehensive characteristic value reception range, indicating that the node can handle various data streams from low to extremely high intensity. A node range with moderate energy efficiency characterization values ​​may only map to a moderate intensity comprehensive characteristic value reception range, indicating that the node is suitable for handling data streams with regular loads. A node range with low energy efficiency characterization values ​​maps to a low comprehensive characteristic value reception range, indicating that the node can only handle data streams with light loads.

[0029] To match node capabilities with reception range, the system iterates through each sensor node and reads its data processing energy efficiency characterization value. Taking a node as an example, assume its calculated data processing energy efficiency characterization value is 85 (this value is a relative scalar value, only used to indicate that it is at a high level). The system compares this value 85 with various data processing energy efficiency characterization value ranges in the information management database. Assume the database defines the range [80, 100] as corresponding to "high processing capability" nodes, and presupposes that the target reception comprehensive characteristic value range for nodes in this range is [50, 200]. Since 85 falls within the [80, 100] range, the system records the corresponding comprehensive characteristic value reception range [50, 200] as the target reception comprehensive characteristic value range for the data acquisition channel of this specific node. This means that, from the perspective of global resource allocation, this node is designed to handle data streams with comprehensive characteristic values ​​between 50 and 200. This matching operation is performed sequentially for all nodes, and each node obtains a target reception comprehensive characteristic value range that matches its processing capability.

[0030] The system then iterates through all sensor types and their combined characteristic values. Assume that in the current system, the combined characteristic value for the accelerometer type is 120, for the gyroscope type it is 75, and for the magnetometer type it is 30. The system matches these combined characteristic values ​​one by one with the range of the aforementioned node (whose receiving range is [50, 200]). The accelerometer's combined characteristic value of 120 falls within the range of [50, 200], therefore the accelerometer is defined as the target receiving sensor type for this node. The gyroscope's combined characteristic value of 75 also falls within this range, so the gyroscope is also defined as the target receiving sensor type for this node. The magnetometer's combined characteristic value of 30 is below the lower limit of 50, therefore the magnetometer is not included in the target receiving type for this node. Through this process, the system concludes that all target receiving sensor types for this node are accelerometers and gyroscopes.

[0031] After determining the target receiving sensor type, the system needs to incorporate all specific data acquisition channels within these types into the cluster. For example, an accelerometer type might contain three independent channels: one for acquiring raw X-axis data, one for acquiring raw Y-axis data, and one for acquiring raw Z-axis data. Similarly, a gyroscope type might contain angular velocity data channels for the X, Y, and Z axes. The system identifies these six specific data acquisition channels (three accelerometer channels and three gyroscope channels) from the overall channel set and categorizes them uniformly as the target receiving data acquisition channels corresponding to that node.

[0032] The system integrates all target data acquisition channels belonging to a given node, forming a logical data stream input set, which is the target data acquisition channel cluster of the sensor. This cluster means that the node is assigned to receive and process data from these six specific channels. For other nodes in the system, the traversal and matching process in steps two through four is repeated. A node with weaker processing power, whose receiving range might only be [20, 60], might only match the magnetometer type (comprehensive feature value 30), thus its target data acquisition channel cluster would only contain the three magnetometer channels. A node with extremely high processing power, whose receiving range might be [40, 250], might match all sensor types, and its cluster would contain all nine channels.

[0033] Example 3: See Figure 4 After the system completes the allocation of the target receiving data acquisition channel cluster, each sensor node is assigned a set of data acquisition channels it is responsible for processing. However, directly processing this raw data may be inefficient, and could even lead to node overload or processing errors due to excessive data volume or inconsistent formats. Therefore, it is necessary to perform personalized preprocessing configuration for each node based on the characteristics of the data inflow.

[0034] The implementation process begins with the evaluation of the data inflow characteristics for each sensor node. For any given sensor node, the system knows that its target data acquisition channel cluster contains several specific data acquisition channels. In Example 1, a basic characteristic value (denoted as ) has been calculated for each data acquisition channel. For the (Several channels), this value quantifies the data acquisition efficiency of that channel. To grasp the overall characteristics of the data flow to this node, the system calculates the average of these basic characteristic values ​​for each channel. This average value is called the data transmission evaluation value of the sensor node (denoted as ). ).

[0035] Its calculation formula is expressed as follows: ; in: This represents the data transmission evaluation value of the current sensor node. This represents the total number of target data acquisition channels allocated to this node. Representing the The basic characteristic values ​​of each data acquisition channel The summation symbol is used. This evaluation value... It provides a single, comprehensive metric that reflects the average strength and rate level of all data streams expected to flow into this node. A high... A lower value means that the node will face a high-throughput, high-load data input environment; a lower value means the node will face a high-throughput, high-load data input environment. The value indicates that the data input is relatively smooth.

[0036] The system needs to determine the preprocessing intensity of the incoming data based on this evaluation value. The system internally presets a data transmission evaluation threshold (denoted as...). This threshold is a key system parameter that delineates the boundary between "regular data flow" and "data flow requiring special attention." Its specific value originates from early-stage system performance analysis and optimization adjustments, aiming to balance processing quality and resource consumption. The formula for calculating the data transmission evaluation threshold T is: ; in, The threshold for data transmission evaluation. This represents the total number of sensor nodes in the system. The average value of the basic characteristic value for assigning channels to the k-th sensor node.

[0037] The decision-making logic is as follows: The system will calculate the node data transmission evaluation value. With preset threshold Compare. If If the current data flow to this node is at a normal or low load level, the system determines that the data flow characteristics are within this range. In this case, the system deems it unnecessary to use a more complex, potentially resource-intensive preprocessing scheme, and therefore decides to continue using the currently applied set of data preprocessing parameters to configure the data received by this node. These parameters may include basic sampling rate filtering, simple outlier removal, or default data format conversion rules. Conversely, if... This indicates that the data flow to the node has reached a high or potentially overloaded level. To ensure that the node can process this data stably and effectively, and to prevent it from becoming a system bottleneck or causing data loss, the system must take more robust preprocessing measures. At this point, the system retrieves and invokes a set of preset, more aggressive data preprocessing parameters from the information management database. These preset parameters are optimized in advance to cope with high-load scenarios and may include: stricter data downsampling rates, more complex data compression algorithms, more aggressive filtering at the expense of some real-time performance for stability, or allocation of more buffer resources. The system uses these new parameters to configure the preprocessing of the data flow received by the node.

[0038] After completing the above preprocessing configuration, each node is ready to receive and process its assigned data stream. At this point, the system generates a clear signal—the data fusion feasible signal. This signal indicates that the data preprocessing stage is complete, and the system status allows and encourages subsequent data fusion operations. The anomaly detection database continuously monitors the system's status signals. Once it receives this data fusion feasible signal, it triggers the next step. It transmits the cluster spatiotemporal information (containing metadata such as node capabilities and data stream characteristics) collected for each sensor node in the previous steps to the designated inspection ports within the anomaly detection database. These inspection ports are the entry points for data into the anomaly detection process, and each port can be considered an independent data receiving and analysis queue. Distributing the cluster spatiotemporal information of different nodes to the inspection ports is a necessary preparation for subsequent independent and parallel anomaly detection of each node's data stream.

[0039] Example 4: See Figure 5The anomaly detection database internally employs multiple detection ports, which are the entities that execute the actual anomaly detection logic. Each detection port generates a series of performance data during runtime, known as port operation data. This data includes: memory utilization, reflecting the memory usage of the port when processing tasks; transmission latency, the time interval between data entering the port and the start of processing; data read / write speed, the speed at which the port accesses storage media to acquire and write data; and data transfer rate, the rate at which the port processes data streams internally. Based on this operational data, the system calculates a detection performance benchmark value using an internal algorithm. This value is a composite indicator used to quantify the current real-time processing capability and efficiency margin of the detection port. A port may have a high absolute processing capability, but if its memory is nearing saturation or latency is high, its detection performance benchmark value will be low, indicating limited currently available detection capability. Supplementary formula for detection performance benchmark value: ; in, To test the performance benchmark value, To detect port memory utilization, This is a normalized value for transmission delay. This is a normalized value for data read / write speed. This is a normalized value for the data transmission rate. is the weighting coefficient, which takes values ​​from 0 to 1 and sums to 1.

[0040] The cluster spatiotemporal information of each sensor node has been transmitted to the inspection port of the anomaly detection database. This cluster spatiotemporal information data mainly includes two core elements: first, the data processing energy efficiency characterization value of the node, which represents the node's inherent processing capability; second, the number of target receiving data acquisition channels allocated to it, which reflects the scale of the data source that the node needs to process. The system comprehensively analyzes this information to calculate a cluster spatiotemporal information evaluation value. This evaluation value aims to quantify the node's overall performance and data processing load in the entire system; a higher value generally indicates that the data flow processed by the node is more critical or more complex.

[0041] The system evaluates the cluster spatiotemporal information of each node and matches it with a corresponding detection performance verification value by querying a pre-defined mapping table. This verification value represents the ideal detection capability level required to successfully process the node's data. The system compares the detection performance benchmark values ​​of all currently available detection ports with the detection performance verification value of a specific node one by one. The goal is to find a port for the node that is both powerful enough (benchmark value higher than verification value) and the most suitable (benchmark value closest to verification value). Once found, this port is designated as the target detection port for the node, and the node's cluster spatiotemporal information data is routed to this target detection port. This process is executed for all nodes to ensure that each node has a dedicated target detection port responsible for the detection task of its data.

[0042] Table 1: Port Performance Matching Table.

[0043]

[0044] Assuming that the calculated cluster spatiotemporal information evaluation value of node Node_A corresponds to a detection performance verification value of 84. As shown in the table above, the baseline value of Port_101, 82, is lower than 84 and does not meet the condition of "higher than". The baseline values ​​of Port_103, 90, and Port_105, 85, are both higher than 84. Among them, the difference between the baseline value of Port_105, 85, and the verification value, 84, is the smallest and closest. Therefore, Port_105 is assigned as the target detection port of Node_A. If the verification value of node Node_B is 76, then both Port_101 and Port_105 meet the condition, but the difference between the baseline value of Port_101, 82, and 76 (6) is greater than the difference between the baseline value of Port_105, 85, and 76 (9). Therefore, Port_101 is "closer" and should be assigned to Node_B. This allocation logic ensures the efficient use of detection resources, avoids resource idleness caused by ports with excessive capacity handling simple tasks, and also prevents delays or errors caused by ports with insufficient capacity handling complex tasks.

[0045] After receiving the specified node cluster spatiotemporal information data at the target detection port, fusion and anomaly detection management are initiated. This port first retrieves the standard spatiotemporal information curve of the multi-channel precision inertial sensor under normal operating conditions from the historical database. This curve is a baseline reference line established in advance through extensive learning from normal data or based on a system theoretical model. Simultaneously, the target detection port receives and plots the real-time cluster spatiotemporal information curve from the sensor node, reflecting the overall state of the data stream at the current moment.

[0046] The algorithm within the port calculates the difference between these two curves in real time, generating a spatiotemporal information deviation curve. This deviation curve visually displays the degree of deviation between the current system state and the ideal normal state. Subsequently, the detection algorithm compares the shape, amplitude, and other characteristics of this real-time deviation curve with a database of preset anomaly pattern features. These preset patterns may include pulse deviation, step deviation, trend drift, etc., each of which may correspond to a sensor or system fault type. If the real-time deviation curve highly matches a preset anomaly pattern feature, the port determines that an anomaly has occurred and generates a preliminary anomaly detection result based on the current deviation curve, including information such as the anomaly type, occurrence time, and deviation amplitude. If the real-time deviation curve fluctuates, but its shape does not match any of the preset anomaly patterns, or the deviation amplitude remains within the allowable noise range, the port will not trigger an anomaly alarm but will continue to monitor the deviation curve. Simultaneously, under this normal state, the port performs fusion management of real-time data based on the standard spatiotemporal information curve, for example, using an adaptive weighted fusion algorithm to output a more stable and reliable fusion data result for use by the upper-level system.

[0047] Example 5: After the target detection port initially determines the existence of an anomaly based on the consistency between the spatiotemporal information deviation curve and the preset anomaly mode characteristics, its work does not end, but enters a more refined analysis stage. The port first needs to determine which known fault or anomaly mode the currently detected anomaly belongs to. The system's preset anomaly mode library stores a variety of typical anomaly mode curves, each corresponding to a physical meaning, such as gyroscope zero-bias mutation, accelerometer scale factor anomaly, intermittent sensor failure, or platform high-frequency micro-vibration, etc. These preset curves are constructed based on historical fault data, simulation analysis, and physical models.

[0048] The analysis process begins with a precise calculation of the matching degree. The port doesn't simply make a qualitative judgment, but rather performs a quantitative analysis of the real-time generated spatiotemporal information deviation curve. Within a time window, it compares the deviation curve point-by-point with each preset abnormal pattern curve in the abnormal pattern library. It calculates the difference between the two at each corresponding data point and aggregates these differences using an algorithm, ultimately obtaining an error value characterizing the overall degree of deviation. This calculation is performed on each preset curve in the pattern library, resulting in a set of error values. Among these error values, the one with the smallest value is identified. This means that among all preset patterns, there is a curve whose shape is most similar to the real-time deviation curve, with the smallest difference. This curve is then identified as the matching abnormal pattern curve.

[0049] To more intuitively express the reliability of the match, the system converts this minimum error value into a matching degree value. The conversion rule ensures that the matching degree value and the error value are inversely correlated; that is, the smaller the error value, the higher the matching degree value, indicating a higher level of confidence in the match. Conversely, the larger the error value, the lower the matching degree value. The matching degree value is a scaled value between 0% and 100%, providing a quantitative measure of the confidence level of the pattern recognition result.

[0050] Based on the identified anomaly pattern curves and the calculated matching degree values, the system generates a structured anomaly detection report. This report is no longer just an alarm indicating "an anomaly exists," but includes detailed information such as the anomaly pattern type, pattern matching degree, timestamp of the anomaly occurrence, and key characteristics of the deviation (e.g., amplitude, rate of change). This report provides system maintenance personnel with accurate diagnostic information and a basis for decision-making in subsequent automated responses.

[0051] The ultimate goal of anomaly detection is not merely to identify problems, but to resolve them or mitigate their impact. Therefore, the system enters a prediction and adjustment phase. Based on the identified anomaly pattern type and combined with standard spatiotemporal information curves (representing the baseline for normal system operation) obtained from historical databases, the system uses predictive algorithms to infer the future behavior trajectory of the multi-channel precision inertial sensors over a short period. This prediction is not blind; it considers how the characteristics of the current anomaly pattern affect the system's dynamic response. For example, if a slow, trend-driven drift anomaly is identified, the prediction algorithm generates a spatiotemporal information prediction curve that gradually deviates from the standard curve; if a sudden step anomaly is identified, the prediction curve will show a stable state at a new level. This prediction curve depicts the state the data might exhibit over a future period without system intervention.

[0052] Based on this spatiotemporal information prediction curve, the system proactively adjusts key parameters in the data fusion process. These adjustments are targeted, aiming to compensate for or suppress the impact of identified anomalies, thereby outputting as reliable data as possible even in the presence of faults. For example, if the prediction indicates that data from a certain axis sensor will drift, the data fusion algorithm may reduce the weight of that sensor's data in the fusion result, instead relying more on data from other normal sensors or data estimated by the algorithm. If the anomaly manifests as increased high-frequency noise, the fusion algorithm may enable stronger high-pass filter parameters. This dynamic parameter adjustment constitutes the system's anomaly response management mechanism, enabling the system to maintain a certain level of functionality and output data reliability even when some units fail or performance degrades, thus improving the system's fault tolerance and overall robustness.

[0053] The entire implementation process embodies a closed-loop management concept, from detection and diagnosis to prediction and final response. It goes beyond simple threshold alarms, using pattern recognition and predictive adjustments to enable multi-channel precision inertial sensor systems to more intelligently respond to internal faults and external interference, ensuring continuous and stable operation in high-precision application scenarios.

[0054] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0055] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A data fusion and anomaly detection method for a multi-channel precision inertial sensor, characterized in that... include: The data acquisition channels connected to the multi-channel precision inertial sensor are statistically analyzed, and the spatiotemporal information of each data acquisition channel is obtained. Based on the spatiotemporal information of each data acquisition channel and the preset sensor node configuration information, the target receiving data acquisition channel cluster of each sensor node is obtained. The spatiotemporal information transmitted by the target receiving data acquisition channel cluster of each sensor node is preprocessed and configured, and the spatiotemporal information of the cluster of each sensor node is statistically analyzed after the preprocessing configuration. The system retrieves and analyzes the running data of each detection port synchronously from the port to be inspected in the anomaly detection database, and integrates the spatiotemporal information of the cluster for anomaly detection management.

2. The data fusion and anomaly detection method for a multi-channel precision inertial sensor according to claim 1, characterized in that, The spatiotemporal information of each data acquisition channel includes the data sampling frequency, data volume, and data transmission rate of each data acquisition channel; The preset sensor node configuration information includes the memory utilization, bandwidth utilization, data processing rate, and CPU utilization of each sensor node.

3. The data fusion and anomaly detection method for a multi-channel precision inertial sensor according to claim 2, characterized in that, The process involves processing the spatiotemporal information from each data acquisition channel and the preset sensor node configuration information. The specific steps are as follows: Based on the spatiotemporal information of each data acquisition channel, the basic feature values ​​of each data acquisition channel are obtained. The basic feature values ​​of each data acquisition channel are used to quantify the data acquisition efficiency of each data acquisition channel. The data acquisition channels are classified according to sensor type to obtain the data acquisition channels corresponding to each sensor type. The average value of the basic feature values ​​of the data acquisition channels corresponding to each sensor type is taken to obtain the comprehensive feature value of the data acquisition channels corresponding to each sensor type. The data processing energy efficiency characterization value of each sensor node is obtained by comprehensively analyzing the preset sensor node configuration information. The data processing energy efficiency characterization value of each sensor node is used to quantify the data processing capability of each sensor node.

4. The data fusion and anomaly detection method for a multi-channel precision inertial sensor according to claim 3, characterized in that, The specific process for obtaining the target received data acquisition channel cluster of each sensor node is as follows: The data processing energy efficiency characterization value of each sensor node is matched with the target reception comprehensive characteristic value range of each data acquisition channel corresponding to the data processing energy efficiency characterization value range stored in the information management database. The target reception comprehensive characteristic value range of each data acquisition channel corresponding to the data processing energy efficiency characterization value of each sensor node is statistically calculated and recorded as the target reception comprehensive characteristic value range of each sensor node's data acquisition channel. The comprehensive characteristic value of the data acquisition channel corresponding to each sensor type is matched with the range of the comprehensive characteristic value of the target reception of the data acquisition channel of each sensor node. If the comprehensive characteristic value of the data acquisition channel corresponding to a certain sensor type is within the range of the comprehensive characteristic value of the target reception of the data acquisition channel of a certain sensor node, then the sensor type is defined as the target reception sensor type of the sensor node. The process is repeated to iterate and count all target reception sensor types of each sensor node and all data acquisition channels corresponding to all target reception sensor types. The data acquisition channels corresponding to all target receiving sensor types of each sensor node are uniformly recorded as the target receiving data acquisition channels of each sensor node, thereby integrating the target receiving data acquisition channel cluster of each sensor node.

5. The data fusion and anomaly detection method for a multi-channel precision inertial sensor according to claim 1, characterized in that, The specific process for preprocessing and configuring the spatiotemporal information transmitted by the target receiving data acquisition channel cluster of each sensor node is as follows: The average value of the basic characteristic values ​​of each target receiving data acquisition channel corresponding to each sensor node is taken to obtain the data transmission evaluation value of each sensor node. The data transmission evaluation value of each sensor node is compared with the set data transmission evaluation threshold. If the data transmission evaluation value of the sensor node is lower than the set data transmission evaluation threshold, the data of the sensor node will continue to be preprocessed and configured with the current data preprocessing parameters. If the data transmission evaluation value of the sensor node is higher than or equal to the set data transmission evaluation threshold, the data of the sensor node will be preprocessed and configured with the preset data preprocessing parameters in the information management database.

6. The data fusion and anomaly detection method for a multi-channel precision inertial sensor according to claim 1, characterized in that, The specific process of statistically analyzing the cluster spatiotemporal information of each sensor node after preprocessing configuration is as follows: After preprocessing and configuration, a data fusion feasibility signal is output. The anomaly detection database receives the data fusion feasibility signal and transmits the cluster spatiotemporal information of each sensor node to the port to be inspected in the anomaly detection database.

7. The data fusion and anomaly detection method for a multi-channel precision inertial sensor according to claim 6, characterized in that, The anomaly detection database synchronously retrieves and analyzes the running data of each detection port from the ports to be inspected. The specific process is as follows: The operating data of each detection port includes the memory utilization, transmission latency, data read / write speed, and data transmission rate of each detection port. Based on the operating data of each detection port, the detection performance benchmark value of each detection port is obtained. The detection performance benchmark value of each detection port is used to quantify the utilization of the detection capability of each detection port. Collect cluster spatiotemporal information data of each sensor node, including the data processing energy efficiency characterization value of each sensor node and the number of target receiving data acquisition channels. The cluster spatiotemporal information data of each sensor node are comprehensively analyzed to obtain the cluster spatiotemporal information evaluation value of each sensor node. The cluster spatiotemporal information evaluation value of each sensor node is used to quantify the comprehensive performance of each sensor node. The detection performance verification value of each sensor node is obtained by matching the cluster spatiotemporal information evaluation value of each sensor node; The detection performance benchmark value of each detection port is compared with the detection performance verification value of each sensor node. If the detection performance benchmark value of a certain detection port is higher than and closest to the detection performance verification value of a certain sensor node, then the detection port is recorded as the target detection port of that sensor node. The target detection ports of each sensor node are obtained by traversing in turn, and the cluster spatiotemporal information of each sensor node is transmitted to the corresponding target detection port.

8. The data fusion and anomaly detection method for a multi-channel precision inertial sensor according to claim 7, characterized in that, The specific process of fusing and managing cluster spatiotemporal information is as follows: Standard spatiotemporal information curves of multi-channel precision inertial sensors are obtained based on historical databases; Real-time cluster spatiotemporal information curves are obtained through the target detection port; The spatiotemporal information deviation curve is obtained based on the real-time cluster spatiotemporal information curve and the standard spatiotemporal information curve. Determine whether the spatiotemporal information deviation curve matches the preset abnormal pattern characteristics. If they match, generate an anomaly detection result based on the spatiotemporal information deviation curve. If they do not match, continue to monitor the spatiotemporal information deviation curve and perform data fusion management based on the standard spatiotemporal information curve.

9. The data fusion and anomaly detection method for a multi-channel precision inertial sensor according to claim 8, characterized in that, The specific process for generating anomaly detection results based on the spatiotemporal information deviation curve is as follows: Analyze the matching degree between the spatiotemporal information deviation curve and the preset anomaly pattern curve to determine the anomaly pattern type; Based on the abnormal mode type and the standard spatiotemporal information curve, the spatiotemporal information prediction curve of the multi-channel precision inertial sensor is obtained. Adjusting data fusion parameters based on spatiotemporal information prediction curves to manage anomaly responses.

10. The data fusion and anomaly detection method for a multi-channel precision inertial sensor according to claim 9, characterized in that, The process of analyzing the matching degree between the spatiotemporal information deviation curve and the preset anomaly pattern curve is as follows: The error value between the spatiotemporal information deviation curve and the curve of each preset anomaly mode is calculated within the time window. Obtain the preset anomaly pattern curve corresponding to the minimum error value, and denote it as the matching anomaly pattern curve; The minimum error value is converted into a matching degree value, and the matching degree value is inversely correlated with the error value. An anomaly detection report is generated based on the matching degree value and the anomaly pattern curve.

Images