A method for functional testing of a sound velocity sensor

Abstract

The application discloses a function test method for a sound velocity sensor, relates to the technical field of acoustic measurement and sensor performance detection, and aims at problems of multiple working conditions being cut off, weak signal detection being difficult and the like in sound velocity sensor testing, constructs a cross-tree connected catalytic heterogeneous decision forest architecture, constructs heterogeneous decision trees for various working conditions, embeds catalytic amplification and data storage in each node, and horizontally interconnects between trees, inputs test signals into each tree in parallel, completes weak signal purification and storage at the node, identifies cross-working condition coupling interference and isolates abnormal data by real-time interaction of error and drift characteristics through horizontal connection branches, outputs three-layer results based on correction of node parameters on the basis of effective data, finally extracts historical data to feed back and iteratively update catalytic parameters, and realizes closed-loop self-optimization. The function test of full-working condition parallel testing and closed-loop self-optimization of the sound velocity sensor is realized.

Description

Technical Field

[0001] This invention relates to the field of acoustic measurement and sensor performance testing technology, specifically to a functional testing method for a sound velocity sensor. Background Technology

[0002] Sound velocity sensors are core sensing devices that indirectly acquire the physical parameters of a medium by measuring the propagation speed of ultrasound waves in it. They are widely used in fields such as marine exploration, petrochemicals, medical ultrasound, industrial non-destructive testing, and fluid measurement. For example, in marine sonar systems, the accuracy of sound velocity measurement directly affects the accuracy of underwater target location and topographic mapping; in chemical production, sound velocity sensors are used for online monitoring of liquid concentration and composition changes; and in the medical field, ultrasonic sound velocity measurement is used for tissue characteristic assessment and image reconstruction.

[0003] Functional testing of sound velocity sensors, namely the systematic verification of their measurement accuracy, repeatability, environmental adaptability, detection blind zone, and fault self-diagnosis capabilities, is a crucial step in ensuring the reliable operation of sensors in practical applications. However, existing functional testing technologies for sound velocity sensors have significant shortcomings: test items are independent, lacking the ability to analyze the correlation between multiple operating conditions. For example, temperature changes not only affect the temperature and pressure adaptability test results but also lead to calibration deviations, reduced repeatability, and expanded blind zones. Existing technologies cannot identify such implicit coupling interference across operating conditions. In low signal-to-noise ratio environments, such as those with bubbly liquids or strong electromagnetic interference, the weak signals output by sound velocity sensors are difficult to detect effectively, and conventional filtering and peak-finding algorithms experience a sharp decline in performance when signal quality is poor. Test data is scattered across different devices or systems, lacking a unified node-level traceability mechanism, making it difficult to quickly locate the root cause once an anomaly occurs. Existing testing methods are mostly open-loop modes with fixed parameters, lacking adaptive optimization capabilities based on historical test data.

[0004] Therefore, in order to address the above problems, there is an urgent need for a functional testing method for sound velocity sensors. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a functional testing method for sound velocity sensors, which solves the problems of difficulty in identifying multi-condition coupling interference and the inability to self-optimize test parameters.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a functional testing method for a sound velocity sensor, comprising the following steps: constructing a cross-tree connected catalytic heterogeneous decision forest architecture, constructing dedicated heterogeneous decision trees for each test condition of the sound velocity sensor, each heterogeneous decision tree embedding a signal catalytic amplification mechanism and a data traceability storage unit, and interconnecting the heterogeneous decision trees through lateral connecting branches; inputting the sound velocity sensor test signal in parallel into each heterogeneous decision tree, using the signal catalytic amplification mechanism to activate and purify weak signals, and storing the processing results in the data traceability storage unit; each heterogeneous decision tree interacts in real time with the error and drift characteristics of other heterogeneous decision trees through the lateral connecting branches, identifying cross-condition coupling interference patterns, and isolating the interfered abnormal data; based on the effective data after isolating the abnormal data, correcting the signal processing parameters and judgment threshold of the current node, each heterogeneous decision tree outputs a single-condition judgment result, and integrates cross-tree correlation features to output a global comprehensive health status, forming a three-layer test result; extracting historical traceability data stored in the data traceability storage unit, feeding back and iteratively updating the parameters of the signal catalytic amplification mechanism, and realizing closed-loop self-optimization of the test architecture.

[0007] Furthermore, the test conditions include calibration conditions, repeatability conditions, temperature and pressure adaptability conditions, blind zone conditions, and fault self-check conditions.

[0008] Furthermore, the cross-tree connected catalytic heterogeneous decision forest architecture includes: five heterogeneous decision trees, each corresponding to a test condition. Each heterogeneous decision tree consists of a root node, intermediate nodes, and leaf nodes arranged vertically. The signal catalytic amplification mechanism and the data traceability storage unit are embedded in the nodes of each heterogeneous decision tree. The root node receives the test signal corresponding to the test condition and distributes the test signal to the intermediate nodes. The intermediate nodes receive the test signal distributed by the root node, perform amplitude compensation and interference suppression processing on the test signal under the action of the signal catalytic amplification mechanism, perform feature splitting determination, and store the processing result and determination result in the data. The data is then distributed to the leaf nodes after being traced back to the storage unit. The leaf nodes are used to receive the processing results and judgment results distributed by the intermediate nodes, store the complete processing link data corresponding to the processing results and judgment results into the data traceability storage unit as historical traceability data for this working condition test, and output the processing results and judgment results as the single working condition judgment results of the heterogeneous decision tree. The heterogeneous decision trees achieve bidirectional data interaction through horizontal connecting branches. The horizontal connecting branches connect the peer nodes of each heterogeneous decision tree and are used to retrieve the status data, historical drift feature data and interference feature data of the current node of other heterogeneous decision trees in real time when each heterogeneous decision tree performs signal judgment and error analysis.

[0009] Furthermore, the three-layer test results specifically include: the first layer consists of the single-condition judgment results output by each heterogeneous decision tree: the calibration condition heterogeneous decision tree outputs the reference sound velocity correction value and zero-point offset error; the repeatability condition heterogeneous decision tree outputs the measurement stability index and fluctuation standard deviation; the temperature and pressure adaptability condition heterogeneous decision tree outputs the temperature drift compensation residual, pressure drift compensation residual, and environmental interference coefficient; the blind zone condition heterogeneous decision tree outputs the minimum measurable distance, weak signal detection threshold, and signal-to-noise ratio; and the fault self-check condition heterogeneous decision tree outputs the fault classification result, signal distortion level, and abnormal alarm level. The second layer consists of cross-condition coupling interference patterns and cross-tree correlation features mined by the real-time interaction of error and drift characteristics of each heterogeneous decision tree through the horizontal connected branches. The cross-condition coupling interference patterns include correlation patterns of calibration offset, repeatability deviation, and blind zone expansion caused by temperature changes. The third layer is the global comprehensive health status output after fusing the single-condition judgment results of each heterogeneous decision tree in the first layer with the cross-tree correlation features in the second layer. The global comprehensive health status includes the comprehensive measurement accuracy level of the sound velocity sensor and the anti-interference robustness score under all operating conditions.

[0010] Furthermore, the signal amplification mechanism includes: configuring a signal amplitude compensation module and an interference suppression module at each node of each heterogeneous decision tree; when the test signal is input to an intermediate node, the signal amplitude compensation module is used to perform amplitude compensation processing on the weak signal in the test signal, and the signal components below the preset detection threshold are amplified to a preset identifiable amplitude range; the interference suppression module simultaneously identifies and filters out bubble clutter interference and electromagnetic clutter interference in the test signal; the test signal after amplitude compensation and interference suppression processing is used as the processing result of the current node and as the judgment basis of the current node.

[0011] Furthermore, the data traceability storage unit is configured at each node of each heterogeneous decision tree to store the following data: the original test signal data received by the current node each time, the processing result data after each signal catalytic amplification mechanism, the feature splitting judgment result data made by the node each time, and the abnormal data identified each time; the data traceability storage unit adds a timestamp label and a node location label to each stored data to realize the full life cycle traceability of test data in the time dimension and the accurate location of the node where the abnormal data is located in the spatial dimension.

[0012] Furthermore, each heterogeneous decision tree interacts with the error and drift characteristics of the other heterogeneous decision trees in real time through the lateral connection branches to identify cross-condition coupling interference patterns. This includes: when performing signal judgment and error analysis for the current condition, each heterogeneous decision tree sends real-time data requests to the other four heterogeneous decision trees through the lateral connection branches; receives current node status data, historical drift characteristic data, and identified interference characteristic data returned by the other heterogeneous decision trees; cross-compares the error characteristics of the current node of this heterogeneous decision tree with the returned data of the other heterogeneous decision trees to identify abnormal fluctuation patterns that occur simultaneously across multiple conditions within the same time window. The abnormal fluctuation patterns include calibration offset, repeatability degradation, and blind zone expansion caused by temperature changes. The identified abnormal fluctuation patterns are defined as cross-condition coupling interference patterns.

[0013] Furthermore, isolating the disturbed abnormal data includes: defining the data in each node of each heterogeneous decision tree that matches the cross-condition coupling interference pattern as abnormal data; marking the storage node corresponding to the abnormal data as a failed node; blocking the data stored in the failed node from participating in the processing and output of subsequent steps; and treating the data stored in each node that is not marked as a failed node as valid data to participate in the signal processing parameter correction and judgment threshold correction of subsequent steps.

[0014] Further, the correction of the signal processing parameters and judgment threshold of the current node includes: using the effective data as a correction benchmark dataset; extracting the signal amplitude features, signal-to-noise ratio features, and error distribution features of each data sample in the correction benchmark dataset; adjusting the signal amplitude compensation coefficient of the current node based on the signal amplitude features, adjusting the interference suppression strength coefficient of the current node based on the signal-to-noise ratio features, and adjusting the judgment benchmark value of the current node based on the error distribution features; and each heterogeneous decision tree performs feature splitting judgment on each intermediately processed signal based on the corrected signal amplitude compensation coefficient, interference suppression strength coefficient, and judgment benchmark value, and outputs the single-condition judgment result.

[0015] Furthermore, historical traceability data is extracted and used to iteratively update the parameters of the signal catalytic amplification mechanism, achieving closed-loop self-optimization of the test architecture. This includes: extracting historical raw test signal data, historical processing result data, and historical feature splitting judgment result data accumulated in the data traceability storage unit of each leaf node, and arranging them according to time series to form the historical performance evolution trajectory of each node; based on the historical performance evolution trajectory, identifying the processing deviation trend of the signal catalytic amplification mechanism of each node under the current parameter configuration; according to the processing deviation trend, iteratively calculating the updated values ​​of the signal amplitude compensation coefficient and interference suppression intensity coefficient of each node, and using the updated values ​​as the input parameters of the signal catalytic amplification mechanism of the corresponding node in the next round of testing; storing the input parameters updated in each round of iterations into the data traceability storage unit of the corresponding node to form a complete record chain of parameter iterative updates.

[0016] The present invention has the following beneficial effects:

[0017] This functional testing method for sound velocity sensors constructs a heterogeneous decision forest architecture with cross-tree connectivity, enabling parallel testing of five operating conditions and significantly improving testing efficiency. Simultaneously, cross-operating condition correlation analysis is achieved through lateral connections between trees, revealing latent coupling interference problems that are undetectable in existing technologies. By embedding signal amplification mechanisms within each node, adaptive amplitude compensation and interference suppression for weak signals are achieved, effectively solving the problem of inaccurate testing of sound velocity sensors in low signal-to-noise ratio environments. Specifically, it identifies and filters out two typical types of interference: bubble clutter and electromagnetic clutter. Furthermore, by configuring data traceability storage units at each node and attaching timestamps and location tags, it achieves… It enables full lifecycle traceability of test data and precise location of abnormal data, solving the problems of scattered and difficult-to-trace test data in existing technologies. Through automatic isolation of abnormal data and failure node marking, it ensures that all data involved in subsequent judgments are valid data, avoiding the contamination of test results by abnormal data. By extracting historical traceability data and iteratively updating node parameters, it achieves closed-loop self-optimization of the test architecture, enabling test accuracy and anti-interference capability to continuously improve with the number of uses. Through a three-layer test result output architecture, it provides both refined diagnostic information for each working condition and a comprehensive global health status assessment of sensors, meeting the different levels of test result requirements under different usage scenarios.

[0018] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0019] Figure 1 This is a flowchart of a functional testing method for a sound velocity sensor according to the present invention.

[0020] Figure 2This is a logic diagram of a functional testing method for a sound velocity sensor according to the present invention.

[0021] Figure 3 This is a flowchart of the closed-loop self-optimizing feedback iteration process in a functional testing method for a sound velocity sensor according to the present invention. Detailed Implementation

[0022] This application embodiment provides a functional testing method for sound velocity sensors, which enables parallel testing of sound velocity sensors under all operating conditions and functional testing with closed-loop self-optimization of the architecture.

[0023] The overall concept of this application's embodiments is as follows:

[0024] For five operating conditions—calibration, repeatability, temperature and pressure adaptability, blind zone, and fault self-check—dedicated heterogeneous decision trees are constructed. Each node embeds a signal catalytic amplification mechanism and a data traceability storage unit. After test signals are input into each tree in parallel, weak signal activation, purification, and storage are completed at each node. Horizontal connecting branches are added between nodes of the same level in each heterogeneous decision tree, enabling each tree to retrieve the error and drift characteristics of other trees in real time during the judgment process, identify cross-operating condition coupling interference patterns, and mark nodes matching abnormal data as failed nodes for isolation. After correcting the parameters of each node based on valid data, three-level results are output: independent subdivision results of a single tree, cross-tree correlation features, and global comprehensive health status. At the same time, historical traceability data is extracted to provide feedback for iterative updates of catalytic parameters, achieving closed-loop self-optimization.

[0025] Please see Figure 1 , Figure 2 , Figure 3 This invention provides a technical solution: a functional testing method for a sound velocity sensor, comprising the following steps:

[0026] Step S1: Construction of a cross-tree connected catalytic heterogeneous decision forest architecture.

[0027] In this embodiment, a cross-tree connected catalytic heterogeneous decision forest architecture is constructed specifically for functional testing of sound velocity sensors. This architecture is not a conventional random forest classifier in machine learning, but rather a signal processing and decision integration system designed specifically for multi-condition testing of sound velocity sensors. Conventional random forests integrate homogeneous trees, with independent trees and no interaction between them; nodes only perform data classification and output single-layer results. The architecture of this invention is a heterogeneous tree architecture, with different tree structures and splitting rules, real-time lateral connectivity between trees, embedded signal catalytic amplification and data traceability storage within nodes, performing signal processing rather than just classification, and outputting three-layer structured results.

[0028] In the specific construction, five dedicated heterogeneous decision trees were built for the five functional test conditions of the sound velocity sensor: calibration condition, repeatability condition, temperature and pressure adaptability condition, blind zone condition, and fault self-check condition. Each heterogeneous decision tree corresponds to one type of condition and has an independent node structure and processing logic, hence the name heterogeneous decision tree. The heterogeneity is reflected in the following aspects: In terms of splitting rules, the calibration tree uses an amplitude deviation threshold comparison type, the repeatability tree uses a fluctuation coefficient variance comparison type, the temperature and pressure tree uses a trend discrimination type, the blind zone tree uses a ratio threshold type, and the fault tree uses a pattern matching type. In terms of the initial value of the compensation coefficient, the calibration tree is set to 2.0, the blind zone tree is set to 8.0, and the others are between 2.0 and 6.0. The interference suppression coefficient is uniformly set to 0.5 and then iterated independently. In terms of judgment thresholds, each tree uses the median standard amplitude, the boundary of 1.5 times the standard deviation of the historical mean, the boundary of ±3% deviation of the standard value, the detection threshold of 1.2 times, and the 50% dividing point between normal and fault waveforms as initial values, and is dynamically updated thereafter.

[0029] Each heterogeneous decision tree consists of three layers vertically: root node, intermediate nodes, and leaf nodes. The root node is located at the top of the heterogeneous decision tree and is the entry point for test signals. The intermediate nodes are located in the middle levels of the heterogeneous decision tree and are the core execution layer for signal processing and judgment. Each heterogeneous decision tree can have multiple levels of intermediate nodes. For example, the calibration condition heterogeneous decision tree has three levels of intermediate nodes, and the blind zone condition heterogeneous decision tree has four levels of intermediate nodes. The leaf nodes are located at the bottom of the heterogeneous decision tree and are the output of the final judgment result and the storage of historical data.

[0030] Each node of each heterogeneous decision tree has two embedded functional units: one is a signal catalytic amplification mechanism, which is used to actively compensate the amplitude and suppress interference of the input test signal; the other is a data traceability storage unit, which is used to persistently store all the data generated by the node in each test.

[0031] In addition to the vertical node hierarchy, horizontal connectivity branches are established between the heterogeneous decision trees. These horizontal connectivity branches are bidirectional data communication links connecting peer nodes in the heterogeneous decision trees. Communication can be implemented using a general serial bus protocol, Ethernet protocol, or a custom internal communication protocol, employing a request-response-acknowledgment sequence. The request includes the sender's tree number, node level, requested data type, and timestamp. The receiver returns a response packet within 50 milliseconds, and the initiator sends an acknowledgment within 100 milliseconds. If no response is received within 200 milliseconds, the request is retransmitted, up to a maximum of three times. Each tree maintains clock synchronization via the NTP protocol, with an error not exceeding 10 milliseconds. Communication anomalies are recorded in the initiator's node storage unit. Through these horizontal connectivity branches, any heterogeneous decision tree can retrieve the current node's status data, historical drift characteristic data, and interference characteristic data of other heterogeneous decision trees in real time when performing signal judgment and error analysis.

[0032] Step S2: Acquisition and parallel input of test signals.

[0033] After the cross-tree connected catalytic heterogeneous decision forest architecture was constructed, functional testing of the sound speed sensor was initiated.

[0034] The sound velocity sensor under test is placed in the corresponding test environment, and the raw ultrasonic echo electrical signals output by the sensor under various operating conditions are collected. These signals are the echo signals received after the ultrasonic waves emitted by the sensor propagate through the medium, carrying sound velocity information. During acquisition, the analog electrical signals are converted into digital signal sequences through an analog-to-digital converter. The sampling rate can be set to collect one million points per second, the quantization precision of each sampling point is 16 bits, and each signal sequence contains 1024 sampling points.

[0035] The acquired raw signals often contain various noises and interferences. Therefore, before inputting them into the heterogeneous decision tree, the signals undergo uniform noise reduction preprocessing. Preprocessing includes three steps: The first step is bandpass filtering, where the passband range is set according to the operating frequency of the sound velocity sensor. For example, if the sensor's operating frequency is 40 kHz, the passband range is set to 35 kHz to 45 kHz to filter out noise outside the passband. The second step is DC bias elimination, where the arithmetic mean of the signal sequence is calculated, and then this mean is subtracted from each sampling point to make the signal symmetrical about the zero axis. The third step is signal normalization, which maps the signal amplitude uniformly to a standard range to facilitate consistent processing of subsequent nodes.

[0036] After preprocessing, the same test signal is simultaneously and in parallel input to the root nodes of the five heterogeneous decision trees. Parallel input means that the five heterogeneous decision trees receive the signal and begin processing simultaneously, rather than processing it sequentially. After receiving the signal, the root node of each heterogeneous decision tree distributes the signal to its subordinate intermediate nodes at each level.

[0037] Step S3: Signal amplification at the node.

[0038] After the signal enters each intermediate node, the signal amplification mechanism embedded in the node begins to work. This mechanism consists of two parts: a signal amplitude compensation module and an interference suppression module.

[0039] The signal amplitude compensation module is responsible for compensating for weak but valid signals in the test signal. Specifically, the module first detects the amplitude of each frequency component in the input signal, identifying signal components with amplitudes below a preset detection threshold as weak signals. The preset detection threshold is set in advance based on the sensor's sensitivity specifications and the background noise level of the test environment. For example, if the sensor's full-scale output amplitude is 200 millivolts and the background noise level is approximately 5 millivolts, the detection threshold can be set to 5% of the full-scale amplitude, i.e., 10 millivolts. Any signal component with an amplitude below 10 millivolts is considered a weak signal requiring compensation.

[0040] After identifying a weak signal, the amplitude compensation module performs amplitude boosting, compensating the amplitude to a preset identifiable range. The upper limit of the preset identifiable amplitude range is the upper limit to avoid signal saturation, and the lower limit is the minimum amplitude to ensure subsequent nodes can effectively identify the signal. For example, the identifiable amplitude range can be set to 20% to 80% of the full-scale amplitude, i.e., 40 mV to 160 mV. The specific compensation method involves determining a gain factor based on the difference between the original amplitude of the signal component and the detection threshold, ensuring the compensated amplitude falls within the identifiable range.

[0041] The interference suppression module works synchronously with the amplitude compensation module. This module is responsible for identifying and filtering out two main types of interference in the test signal: one is bubble clutter interference, which presents as sudden spikes in the time domain and is typically distributed in the 10 kHz to 50 kHz range in the frequency domain; the other is electromagnetic clutter interference, including power frequency interference and its harmonics, as well as high-frequency switching noise. The harmonics are at 50 Hz, 100 Hz, 150 Hz, etc., and the high-frequency switching noise is above 1 MHz. The interference suppression module uses digital filtering algorithms to identify and filter out these interferences. Specifically, it may use bandpass filters to retain the effective signal bandwidth, notch filters to suppress specific frequency interference, or wavelet denoising algorithms to adaptively suppress non-stationary noise.

[0042] The signal after amplitude compensation and interference suppression processing serves as the effective processing result of the current node. It is stored in the data traceability storage unit of the current node, serves as the basis for the node to perform feature splitting determination, and is also distributed to the next level node for further processing.

[0043] The amplitude compensation module employs a digital programmable gain amplifier combined with a gain control algorithm. Multiple sets of gain coefficients are pre-stored within the module. When a signal component is detected to be below the detection threshold, the difference is calculated, and the corresponding gain factor is selected from the gain coefficient table: the larger the difference, the higher the gain factor. The gain coefficient table is pre-generated based on the sensor model during system initialization. The interference suppression module uses a three-stage cascaded filtering system: the first stage is a bandpass filter with a passband of ±5 kHz from the sensor's operating frequency; the second stage is an adaptive notch filter that automatically tracks 50 Hz and its harmonics and generates corresponding notches; the third stage is wavelet thresholding denoising, using Daubechies wavelets, 4-level decomposition, and soft thresholding. After the three stages are cascaded, the signal-to-noise ratio is typically improved by 15 to 25 dB.

[0044] Step S4: The working mechanism of the data traceability storage unit.

[0045] Each node in each heterogeneous decision tree is equipped with a data traceability storage unit. This storage unit can be a non-volatile flash memory chip, a dedicated storage area in random access memory, or an interface connected to an external database.

[0046] Whenever a node receives a signal, completes catalytic amplification, or makes a feature splitting determination, it stores the relevant data in the data traceability storage unit. The stored data includes four categories: the first category is raw test signal data, which is the unprocessed signal waveform sequence received by the node each time; the second category is processing result data, which is the intermediate result data after the node performs signal catalytic amplification processing each time; the third category is feature splitting determination result data, which is the determination conclusion made by the node each time; and the fourth category is abnormal data, which is the signal or determination result that the node identifies as abnormal each time.

[0047] To facilitate subsequent traceability, the data traceability storage unit attaches two tags to each piece of stored data: one is a timestamp tag, which records the precise time when the data was stored, accurate to milliseconds, for example: June 21, 2026, 14:35:12.300; the other is a node location tag, which records the heterogeneous decision tree number, node type, and node level number to which the data belongs, for example: second tree - third-level intermediate node.

[0048] With these two tags, when it is necessary to trace a certain abnormal data later, it is possible to accurately locate when the data was generated, which node of which heterogeneous decision tree, and achieve dual tracing in both time and space dimensions.

[0049] Step S5: Identification of the lateral connectivity branches and cross-condition coupling interference patterns.

[0050] During the signal judgment and error analysis process, each heterogeneous decision tree interacts with other heterogeneous decision trees in real time through lateral connection branches to identify cross-condition coupling interference patterns.

[0051] The specific interaction process is as follows:

[0052] When the current node of a heterogeneous decision tree is performing a signal determination, it sends a real-time data request to the sibling nodes of the other four heterogeneous decision trees through a lateral connection branch. The request includes the current time window identifier and the required data type.

[0053] After receiving the request, the nodes of the other four heterogeneous decision trees return their current node state data, historical drift feature data, and identified interference feature data to the requester through the lateral connection branches.

[0054] The requesting node cross-compares its current error characteristics with the returned data from other nodes. The comparison dimensions include three aspects: whether the trend of signal amplitude change is consistent, whether the direction of the deviation in the judgment result is the same, and whether the time of deviation occurrence is synchronized.

[0055] If multiple operating conditions exhibit synchronous abnormal fluctuations within the same time window, such as within the same test batch or the same environmental change cycle, these synchronously occurring abnormal fluctuation patterns are identified as cross-operating condition coupling interference patterns. The three criteria for determining cross-operating condition coupling interference patterns specifically include: a synchronous time window error tolerance of ±1 second; at least two types of operating conditions exhibiting abnormalities simultaneously, which are considered coupling interference, while single-tree abnormalities are considered independent noise; and anomaly determination is defined as the current error exceeding the mean of the most recent 100 tests of that node by ±3 times the standard deviation, and lasting for more than two consecutive sampling cycles, with the mean and standard deviation being updated on a rolling basis. Fluctuations that do not simultaneously meet the above three conditions are not included in coupling interference.

[0056] The sampling period refers to the total duration of a single complete test process, from signal input to a node to the completion of signal amplification and feature splitting determination, and output of the result to the next level node. The duration of each sampling period is equal to the sum of the signal processing time and determination time of that node. Depending on the node level and operational complexity, the duration of a single sampling period ranges from 50 milliseconds to 200 milliseconds. The root node only performs signal distribution, with a minimum period of approximately 50 milliseconds; intermediate nodes perform amplification and splitting determination, with periods of approximately 100 to 150 milliseconds; and leaf nodes perform final determination and data storage, with periods of approximately 150 to 200 milliseconds. The sampling period duration of all nodes is determined through actual measurement and calibration during system initialization and stored in the configuration parameters of each node. The sampling periods of nodes at each level within the same tree are independent of each other.

[0057] For example, when the ambient temperature rises, the zero-point offset error of the calibration condition heterogeneous decision tree increases, the fluctuation standard deviation of the repeatability condition heterogeneous decision tree increases, and the minimum measurable distance of the blind zone condition heterogeneous decision tree increases. If these three anomalies occur simultaneously within the same time window, they are identified as cross-condition coupling interference caused by temperature changes. This pattern is recorded and stored in the data traceability storage unit of each node for subsequent retrieval.

[0058] Step S6: Isolation of abnormal data and marking of failed nodes.

[0059] After identifying the cross-operating condition coupling interference pattern, each heterogeneous decision tree isolates abnormal data according to the pattern.

[0060] The specific approach is as follows: data in each node that matches the cross-condition coupling interference pattern is defined as anomalous data. The matching criteria include three conditions: the data amplitude exceeds the range of the historical data mean of that node plus or minus three standard deviations; the data trend is consistent with the identified coupling interference pattern; and the timestamp of the data differs from the timestamp of the anomalous data that appears synchronously in other heterogeneous decision trees by no more than one second. Data that meets any of the above conditions is identified as anomalous data.

[0061] The storage node corresponding to the data identified as abnormal is marked as a failed node. The marking method is to write a failure status identifier into the data traceability storage unit of that node.

[0062] Once a node is marked as invalid, all data stored in that node is blocked and will no longer participate in any subsequent processing or output. Specifically, a judgment logic is set at the beginning of the data processing flow of each node: before each processing step, the status flag of the current node is checked; if it is invalid, the data reading and participation in subsequent calculations of that node are skipped.

[0063] The logic for blocking and clearing data at failed nodes is specifically a three-level process: A valve flag is set at the input; when a node fails, new data is rejected, and the parent node skips the node and directly distributes the data to the next level. Cache clearing: Within 500 milliseconds after a failure, the temporary cached data of the node is cleared, while the historical data is retained only by marking it as failed. Blocking is permanent and does not have an automatic release mechanism; administrators must verify and eliminate the root cause of the anomaly before manually restoring the data. After restoration, the data will rejoin the judgment process in the next round. During the failure period, the data retains the failure mark and is not considered valid data.

[0064] Data stored in nodes that are not marked as failed nodes are considered valid data and continue to participate in subsequent signal processing parameter correction and judgment threshold correction.

[0065] Step S7: Correction of signal processing parameters and judgment threshold.

[0066] After isolating abnormal data, the signal processing parameters and judgment thresholds of the current node are corrected based on the remaining valid data.

[0067] The valid data stored in all nodes that were not marked as invalid are aggregated to form a corrected baseline dataset; three types of features are extracted from the corrected baseline dataset:

[0068] The first category is signal amplitude characteristics. The extraction method involves calculating three statistics for the waveform sequence of each data sample: maximum value, arithmetic mean, and root mean square value. For example, if a data sample has a peak value of 120 millivolts, a mean of 50 millivolts, and an effective value of 70 millivolts, these are the signal amplitude characteristics of that sample.

[0069] The second type is the signal-to-noise ratio (SNR) feature. The extraction method involves first identifying the effective signal range and the noise range in the data sample, and then calculating the ratio of the effective signal power to the noise power. For example, if the effective signal power of a data sample is 100 milliwatts and the noise power is 1 milliwatt, then the SNR is 100:1, which translates to a decibel value of 20. A higher SNR indicates better signal quality.

[0070] The third category is error distribution characteristics. The extraction method is to subtract the measured value of each data sample from the standard value to obtain the error value. Then, four statistics are calculated for all error values: the mean of the error, the standard deviation of the error, the skewness of the error, and the kurtosis of the error. The mean reflects systematic deviation, the standard deviation reflects the degree of random fluctuation, the skewness reflects the asymmetry of the error distribution, and the kurtosis reflects the sharpness of the error distribution.

[0071] After extracting the above three types of features, they are used to adjust the three parameters respectively:

[0072] The first parameter is the signal amplitude compensation coefficient, which controls the gain of the amplitude compensation module. Its value range is generally from 1 to 100. The adjustment method is as follows: calculate the average signal amplitude of all samples in the correction reference dataset, compare it with the preset target amplitude, and if the average amplitude is lower than the target value, increase the compensation coefficient; if it is higher than the target value, decrease the compensation coefficient.

[0073] The second parameter is the interference suppression strength coefficient, which controls the filtering strength of the interference suppression module. Its value typically ranges from 0 to 1. The adjustment method is as follows: calculate the average signal-to-noise ratio (SNR) of all samples in the corrected benchmark dataset. If the average SNR is lower than the preset target value, such as 20 dB, increase the suppression strength; if it is higher than the target value, decrease the suppression strength.

[0074] The third parameter is the decision benchmark value, which is the threshold used when a node performs feature splitting. The adjustment method is as follows: based on the mean and median of the error distribution features in the corrected benchmark dataset, the decision benchmark value is adjusted to the center position of the error distribution.

[0075] Each heterogeneous decision tree processes and determines subsequent input signals based on the corrected amplitude compensation coefficient, interference suppression strength coefficient, and judgment benchmark value.

[0076] Step S8, the specific execution of feature splitting determination.

[0077] After completing the signal amplification process, each node performs feature splitting determination. Feature splitting determination refers to the process by which a node divides the data into different branches based on the currently processed signal data and according to preset feature attributes and determination benchmark values.

[0078] The specific steps are as follows:

[0079] S8.1 Extract the feature attribute values ​​of the current signal data. Extractable feature attributes include signal amplitude, signal frequency, signal phase, signal-to-noise ratio, etc. Different nodes may extract different feature attributes. For example, nodes in calibration conditions focus more on amplitude characteristics, while nodes in blind zone conditions focus more on signal-to-noise ratio characteristics.

[0080] S8.2, comparing the extracted feature attribute value with the judgment reference value of the current node. The comparison is performed by calculating the difference between the feature attribute value and the judgment reference value.

[0081] S8.3, distributing data to the corresponding lower-level nodes according to the comparison result. For example, if the feature attribute value is greater than the judgment reference value, the data is distributed to the left branch; if the feature attribute value is less than or equal to the judgment reference value, the data is distributed to the right branch. Each level of node performs such a judgment once, and the signal is transmitted downward step by step along the longitudinal level of the heterogeneous decision tree until it reaches the leaf node.

[0082] The feature attributes of each node are determined according to working conditions: the calibration working condition takes signal amplitude and first wave arrival time as main features; the repeatability working condition takes amplitude fluctuation and time jitter as features; the temperature and pressure adaptability working condition takes amplitude temperature drift rate and time delay pressure sensitivity as features; the blind area working condition takes signal-to-noise ratio and the amplitude ratio of secondary wave to first wave as features; the fault self-checking working condition takes waveform distortion rate and spectral energy offset as features. The initial value of the judgment reference value is set by statistical learning: no less than 50 groups of standard signal samples are pre-collected for each type of working condition, the mean value and standard deviation of each feature attribute are calculated, and mean value ± 1.5 times standard deviation is taken as the initial reference range.

[0083] The calculation objects of the above mean value and standard deviation are pre-collected standard signal samples. The specific method is as follows: under the factory calibration state or known intact state of the sensor, no less than 50 groups of standard test signal samples are collected respectively for each type of working condition, and each group of samples contains all feature attribute values corresponding to the working condition. The arithmetic mean value and standard deviation are calculated respectively for each feature attribute, the mean value plus 1.5 times standard deviation is taken as the upper limit of the judgment reference value, and the mean value minus 1.5 times standard deviation is taken as the lower limit of the judgment reference value. If a certain feature attribute is of positive correlation type, the judgment reference value is the mean value minus 1.5 times standard deviation; if it is of negative correlation type, the judgment reference value is the mean value plus 1.5 times standard deviation. In the initial judgment, those with feature attribute values better than the reference value are judged as qualified, and those worse than the reference value are judged as abnormal.

[0084] Step S9, generation and output of three-layer test results.

[0085] After the leaf nodes of each heterogeneous decision tree receive the final judgment signal, they output the single working condition judgment result of the corresponding working condition, which forms the first layer of the three-layer test result.

[0086] The specific output content of each working condition in the first layer includes:

[0087] The calibration condition heterogeneous decision tree outputs two parameters: the reference sound velocity correction value, which is the difference between the sensor's measured sound velocity value and the standard sound velocity value, in meters per second. For example, if the measured value is 1498 meters per second and the standard value is 1500 meters per second, then the correction value is -2 meters per second; and the zero-point offset error, which is the DC bias of the sensor output when there is no signal input, in millivolts. For example, the zero-point offset is 2.5 millivolts.

[0088] The heterogeneous decision tree for repetitive operating conditions outputs two parameters: measurement stability index, which is the ratio of the maximum deviation to the average value of multiple repeated measurements, expressed as a percentage. For example, if the maximum deviation of ten measurements is 3 meters per second and the average value is 1500 meters per second, then the stability index is 2.0%; and fluctuation standard deviation, which is the standard deviation of multiple measurements, in meters per second. For example, the standard deviation is 0.8 meters per second.

[0089] The heterogeneous decision tree for temperature and pressure adaptability outputs three parameters: temperature drift compensation residual, which is the residual deviation remaining after temperature compensation, in meters per second per degree Celsius, for example, 0.02 meters per second per degree Celsius; pressure drift compensation residual, which is the residual deviation remaining after pressure compensation, in meters per second per megapascal, for example, 0.05 meters per second per megapascal; and environmental interference coefficient, which is a dimensionless comprehensive coefficient reflecting the comprehensive influence of environmental factors such as temperature, pressure, and medium density on measurement accuracy. The value range is generally from 0 to 1, and the smaller the value, the stronger the resistance to environmental interference, for example, 0.15.

[0090] The heterogeneous decision tree for blind zone conditions outputs three parameters: minimum measurable distance, which is the shortest distance at which the sensor can reliably detect reflected sound signals, in millimeters or centimeters, for example, 5 centimeters; weak signal detection threshold, which is the minimum signal amplitude that the sensor can reliably identify, in millivolts, for example, 8 millivolts; and signal-to-noise ratio, which is the ratio of the effective signal amplitude to the noise amplitude, expressed in decibels, for example, 25 decibels.

[0091] The fault self-check heterogeneous decision tree outputs three parameters: fault classification result, i.e. the identified fault type identifier, such as sensor failure, signal channel open circuit, sensitivity drop of more than 20%, etc.; signal distortion level, divided into three levels: normal, slight distortion, and severe distortion; and abnormal alarm level, divided into four levels: normal, attention, warning, and danger.

[0092] The second layer of test results reveals cross-condition coupling interference patterns and cross-tree correlation features. Specifically, it involves the extraction of cross-condition coupling interference pattern parameter sets and cross-tree correlation feature vectors from heterogeneous decision trees through real-time interaction of their respective error and drift features via lateral connecting branches. For example, the correlation pattern of calibration offset, repeatability deviation, and blind zone expansion caused by temperature changes is a typical example of the second layer's output.

[0093] The second-layer cross-condition coupling interference parameter set includes three quantitative indicators: coupling influence factor, associated fluctuation amplitude, and coupling response time. The coupling influence factor, ranging from 0 to 1, reflects the intensity of interference from a certain environmental factor to other conditions. The associated fluctuation amplitude represents the average amplitude of synchronous fluctuations across all conditions, and the coupling response time represents the time difference between the change in the environmental factor and the occurrence of synchronous fluctuations. The cross-tree association feature vector is a five-dimensional vector, with each dimension corresponding to the current comprehensive state score of a decision tree. Qualitative grading: a coupling influence factor less than 0.3 indicates weak coupling, 0.3 to 0.6 indicates medium coupling, and greater than 0.6 indicates strong coupling; associated fluctuation amplitude less than 10% of normal fluctuation indicates slight impact, 10% to 30% indicates moderate impact, and more than 30% indicates severe impact.

[0094] The third-layer test result is the overall global health status. Specifically, it is generated by fusing the single-condition judgment results of each heterogeneous decision tree in the first layer with the cross-tree association features in the second layer, and using a weighted comprehensive evaluation method to obtain two final indicators: the overall measurement accuracy level of the sound velocity sensor and the all-condition anti-interference robustness score. The overall measurement accuracy level is divided into four levels: Level 1 is excellent, Level 2 is good, Level 3 is acceptable, and Level 4 is unacceptable. The all-condition anti-interference robustness score ranges from 1 to 100 points, with higher scores indicating stronger anti-interference capabilities, for example, 85 points.

[0095] The overall measurement accuracy level of the sound velocity sensor is evaluated using a weighted comprehensive assessment method. The specific steps include: assigning weights to the output parameters of each operating condition in the first layer. The principle for setting the weights is: the parameter with the greater impact on the overall performance of the sensor, the higher the weight. For example, the reference sound velocity correction value in the calibration condition directly determines the sensor's measurement accuracy, with a weight of 30%; the fluctuation standard deviation in the repeatability condition reflects the sensor's measurement consistency, with a weight of 20%; the temperature and pressure drift compensation residuals in the temperature and pressure adaptability condition reflect the sensor's environmental adaptability, with a weight of 20%; the minimum measurable distance in the blind zone condition reflects the sensor's detection capability, with a weight of 15%; and the fault classification and alarm level in the fault self-check condition reflect the sensor's reliability, with a weight of 15%. The sum of all weights is 1. The actual measured value of each parameter is then scored. The scoring is based on comparing the measured value with a preset standard value or acceptable range. For example, the standard for the reference sound speed correction value is zero. A deviation within ±1 meter per second is rated 100 points, within ±2 meters per second 80 points, within ±3 meters per second 60 points, and a deviation exceeding ±3 meters per second 40 points. The scores for each parameter are multiplied by their corresponding weights and then summed to obtain the overall score. Overall scores of 90 points or above are rated Level 1, 80 to 90 points Level 2, 60 to 80 points Level 3, and below 60 points Level 4.

[0096] Supports scenario-based weight adjustment: The system can preset underwater detection weights as follows: 35, 15, 25, 20, 5; industrial pipeline weights as follows: 20, 25, 15, 10, 30; and medical ultrasound weights as follows: 30, 20, 20, 15, 15. Users select these weights during initialization. Every 100 rounds, the system calculates the anomaly frequency for each tree. If the anomaly frequency for a certain operating condition exceeds twice the average, the weight for that condition is automatically increased by 5%; if it is below 50% of the average, it is decreased by 3%. A single adjustment cannot exceed 5%, and the cumulative adjustment cannot exceed ±50% of the initial value. Manual configuration has higher priority than automatic configuration.

[0097] The full-condition interference robustness rating also employs a comprehensive evaluation method, but with a different focus. This rating primarily reflects the sensor's ability to resist interference under various operating conditions. The rating is based on three dimensions: the first is the improvement in signal-to-noise ratio after interference suppression under each operating condition; a larger improvement indicates stronger interference resistance. The second is the frequency of abnormal data occurrences; a lower frequency indicates stronger interference resistance. The third is the suppression effect of cross-condition coupling interference, i.e., after identifying the coupling interference pattern and adjusting parameters, whether the coupling interference is effectively suppressed in subsequent tests. The scores for each dimension are also weighted and summed to arrive at a comprehensive score ranging from 0 to 100. This rating can be directly used to compare the interference resistance of different sensors, and can also be used to track the performance of the same sensor in different batches of tests.

[0098] Step S10: Extraction and closed-loop self-optimization of historical traceability data.

[0099] After completing a round of testing, the system extracts historical data accumulated in the traceability storage unit from each leaf node, which is used to iteratively update the parameters of the signal catalytic amplification mechanism to achieve closed-loop self-optimization.

[0100] The specific steps are as follows:

[0101] S10.1 Extract the three types of historical data stored in each leaf node: raw test signal data, processed result data, and feature splitting judgment result data. Arrange these data in chronological order to form the historical performance evolution trajectory of each node. The horizontal axis of the evolution trajectory is the test time or test batch number, and the vertical axis is the key performance indicators of each node, such as the change curve of the signal amplitude compensation coefficient, the change curve of the interference suppression intensity coefficient, and the change curve of the judgment accuracy.

[0102] S10.2, Based on historical performance evolution trajectories, identify the processing deviation trends of each node under the current parameter configuration. The identification method is as follows: calculate the slope of the evolution trajectory. If the slope is consistently positive, it indicates that the indicator is continuously rising, possibly indicating an overcompensation trend; if the slope is consistently negative, it indicates that the indicator is continuously declining, possibly indicating an undercompensation trend. Simultaneously, calculate the cumulative deviation between the evolution trajectory and the preset target value. If the cumulative deviation exceeds a preset threshold, for example, exceeding 10% of the target value, it indicates that parameter adjustments are needed.

[0103] S10.3, based on the identified processing deviation trend, iteratively calculate the updated values ​​of the signal amplitude compensation coefficient and interference suppression strength coefficient for each node. The iterative calculation adopts a stepwise approximation method: in each iteration, based on the direction and magnitude of the current deviation, add a small increment opposite to the direction of the deviation to the coefficient value at the current position. The magnitude of the increment is controlled by the learning rate, which can be set to 0.05, meaning that the adjustment amplitude each time does not exceed 5% of the current value. Through multiple iterations, the coefficient value gradually approaches the optimal value.

[0104] The iterative calculation employs a dual convergence criterion: First, convergence is considered achieved and iteration stops when the change in the updated value for three consecutive iterations is less than 1% of the current value. Second, when the number of iterations reaches a preset maximum, such as 50, the system forcibly stops regardless of whether the first condition is met, preventing infinite loops. After each iteration, the system compares the absolute difference between the current and previous updated values. If the absolute difference divided by the current value is less than 0.01, a small change count is accumulated. After three consecutive small changes are accumulated, a convergence signal is triggered.

[0105] When the parameter values ​​updated in each iteration are stored in the data traceability storage unit of the corresponding node, the convergence status and final update value of that iteration are also recorded, forming a complete parameter evolution record chain. When the next test begins, the system defaults to loading the parameter values ​​converged in the previous round as the initial values. If the previous round was forcibly stopped due to reaching the maximum number of iterations, the learning rate will be halved and the iteration will restart in this round to ensure eventual convergence.

[0106] In step S10.4, the updated values ​​obtained from the iterative calculation are used as input parameters for the signal catalytic amplification mechanism of the corresponding node in the next round of testing. In other words, in the next round of testing, the node will use the updated amplitude compensation coefficient and interference suppression strength coefficient to perform signal processing.

[0107] In step S10.5, the parameter values ​​updated in each iteration are stored in the data traceability storage unit of the corresponding node, forming a complete record chain of parameter iteration updates together with historical data. In this way, the entire test architecture forms a closed loop of testing, storage, analysis, adjustment, and retesting, which can continuously optimize itself as the number of tests increases.

[0108] To enable those skilled in the art to better implement this invention, specific setting examples of each main preset parameter are given below. It should be noted that the following values ​​are merely examples, and in actual applications, they can be adjusted accordingly based on the specific sensor model, testing environment, and accuracy requirements.

[0109] Preset detection threshold: Set to 5% of the sensor's full-scale output amplitude. If the sensor's full-scale output is 100 millivolts, then the detection threshold is 10 millivolts. This threshold can be adjusted within the range of 3% to 10% based on the background noise level. The higher the background noise, the higher the threshold should be to avoid false triggering.

[0110] The preset identifiable amplitude range is set as follows: the lower limit is 20% of the full-scale amplitude, and the upper limit is 80% of the full-scale amplitude. For a 200 mV full-scale sensor, the identifiable range is 40 mV to 160 mV. The lower limit of this range ensures that the compensated signal can be effectively identified by subsequent nodes, while the upper limit prevents signal saturation and distortion.

[0111] Target signal-to-noise ratio (SNR): Set to 20 dB. When the average SNR is below 20 dB, increase the interference suppression strength; when it is above 20 dB, appropriately decrease the suppression strength to retain more signal details.

[0112] Anomaly detection threshold: set at the mean plus or minus three standard deviations. Data exceeding this range is considered potentially outlier. The choice of three standard deviations is based on the 30 rule of normal distribution in statistics, meaning that data exceeding three standard deviations has a probability of less than 0.3%, which is considered a low-probability event and can be identified as anomaly.

[0113] Learning rate: set to 0.05. The parameter adjustment increment during each iteration should not exceed five percent of the current value to ensure the smoothness of the iteration process and avoid drastic parameter oscillations.

[0114] Time window: set to one second. When identifying cross-condition coupling interference patterns, synchronous abnormal fluctuations occurring within the same second in each heterogeneous decision tree node are considered correlated.

[0115] Number of repeated tests: set to ten. In the repetitive operating condition test, ten repeated measurements are performed, and the stability index and standard deviation of fluctuation are calculated.

[0116] Temperature test points: Five temperature points are set, namely 0 degrees Celsius, 10 degrees Celsius, 20 degrees Celsius, 30 degrees Celsius and 40 degrees Celsius, covering the normal operating temperature range of the sensor.

[0117] Pressure test points: Four pressure points are set, namely 0.1 MPa, 0.5 MPa, 1.0 MPa and 2.0 MPa, covering the normal operating pressure range of the sensor.

[0118] The above preset parameter examples are applicable to a conventional piezoelectric sound velocity sensor with a full-scale range of 200 mV and an operating frequency of 40 kHz. When the sensor model changes, the parameters are adjusted according to the following rules: The preset detection threshold is always set to 5% of the sensor's full-scale amplitude. The full-scale amplitude is obtained by reading the sensor datasheet or through actual measurement and calibration. The larger the full-scale amplitude, the higher the absolute value of the detection threshold. The preset recognizable amplitude range is always set to 20% to 80% of the full-scale amplitude. The lower limit ensures that the amplified signal is not overwhelmed by subsequent noise, and the upper limit avoids saturation distortion caused by approaching the full scale. The signal-to-noise ratio target value is set according to the sensor application scenario: 15 dB for underwater long-distance detection scenarios and 25 dB for short-range high-precision measurement scenarios. The learning rate is adjusted according to the test sample size: 0.03 for samples smaller than 100 batches to ensure stability, and 0.08 for samples larger than 500 batches to accelerate convergence. The time window is set according to the rate of environmental change: 0.5 seconds for sudden temperature changes and 2 seconds for slow changes.

[0119] The initial values ​​of each parameter are obtained through air test calibration before the system is first run: the sensor is placed in a standard test environment, 100 sets of background signals are collected, their amplitude statistical distribution is calculated, and the mean plus 3 times the standard deviation is set as the initial value of the detection threshold; 100 sets of standard signals are collected, their amplitude distribution range is calculated, and 80% of the lower limit of the range is set as the lower limit of the identifiable amplitude range, and 120% of the upper limit is set as the upper limit of the identifiable amplitude range.

[0120] The following is a timing description of a complete test process:

[0121] S11.1 Construct a cross-tree connected catalytic heterogeneous decision forest architecture. This step is completed before the first test, and the constructed architecture can be reused in subsequent tests.

[0122] S11.2, Collect the original ultrasonic echo signals of the sound velocity sensor under five working conditions and perform unified noise reduction preprocessing.

[0123] S11.3, the preprocessed test signals are input in parallel to the root nodes of the five heterogeneous decision trees, and each root node distributes the signals to the intermediate nodes at each level.

[0124] S11.4, each intermediate node performs signal catalytic amplification processing and stores the processing results in the data traceability storage unit.

[0125] S11.5, each intermediate node performs feature splitting determination, stores the determination result in the data traceability storage unit and then distributes it to the next level node, until the leaf node.

[0126] S11.6, each heterogeneous decision tree identifies cross-operating condition coupling interference patterns by using real-time interaction errors and drift characteristics of horizontally connected branches.

[0127] S11.7 Based on the identified coupling interference patterns, the storage nodes corresponding to the abnormal data are marked as failed nodes, and the abnormal data is isolated.

[0128] S11.8, based on the effective data after isolation, corrects the signal processing parameters and judgment thresholds of each node.

[0129] S11.9: Each leaf node outputs the single-condition judgment result, and at the same time, it integrates cross-tree correlation features to output the global comprehensive health status.

[0130] S11.10 Extract the historical traceability data stored in each node, iteratively update the parameters of the signal catalytic amplification mechanism, and prepare for the next round of testing.

[0131] The above ten steps constitute a complete test loop. As the number of test rounds increases, the parameters of each node are continuously optimized, and the performance of the test architecture continues to improve.

[0132] In summary, this application has at least the following effects:

[0133] By enabling cross-condition correlation analysis through horizontally connected branches, the problem of implicit coupling interference caused by factors such as temperature and pressure simultaneously across multiple conditions cannot be identified by conventional independent item-by-item testing. By embedding a catalytic amplification mechanism within nodes, adaptive compensation and filtering of bubble clutter and electromagnetic clutter are performed, significantly improving the reliability of weak signal detection in low signal-to-noise ratio environments. By adding timestamps and location tags to node-level data traceability storage units, full lifecycle traceability of test data and precise anomaly location are achieved. After an abnormal node is marked as invalid, its data is blocked from participating in subsequent judgments, ensuring the purity of valid data. By iteratively updating node parameters through historical data feedback, the test architecture has closed-loop self-optimization capabilities, and its accuracy and anti-interference ability continue to improve with each test round. Through a three-layer output architecture, refined diagnosis of each condition and comprehensive global health assessment are provided simultaneously, meeting the multi-level testing needs in different scenarios.

[0134] Those skilled in the art will understand that embodiments of the present invention can be provided as methods. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0135] This invention is described with reference to a flowchart of a method according to embodiments of the invention. It should be understood that the combination of each step in the flowchart can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing device, generate instructions for implementing the process. Figure 1 A device for a function specified in one or more processes.

[0136] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 The function specified in one or more processes.

[0137] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 Steps of a specified function in one or more processes.

[0138] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the invention.

[0139] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for functional testing of a sound velocity sensor, characterized in that, Includes the following steps: A cross-tree connected catalytic heterogeneous decision forest architecture is constructed. Dedicated heterogeneous decision trees are built for the test conditions of sound speed sensors. Each heterogeneous decision tree embeds a signal catalytic amplification mechanism and a data traceability storage unit. The heterogeneous decision trees are interconnected through horizontal connecting branches. The sound velocity sensor test signal is input into each heterogeneous decision tree in parallel. The weak signal is activated and purified by the signal catalytic amplification mechanism, and the processing result is stored in the data traceability storage unit. Each heterogeneous decision tree interacts with the error and drift characteristics of other heterogeneous decision trees in real time through the horizontal connection branches, identifies the cross-operating condition coupling interference pattern, and isolates the abnormal data affected by interference; Based on the valid data after isolating abnormal data, the signal processing parameters and judgment threshold of the current node are corrected, each heterogeneous decision tree outputs the single-condition judgment result, and the cross-tree correlation features are integrated to output the global comprehensive health status, forming a three-layer test result; Extract historical traceability data stored in the data traceability storage unit, feed back and iterate the parameters of the signal catalytic amplification mechanism, and realize closed-loop self-optimization of the test architecture.

2. The functional testing method for a sound velocity sensor according to claim 1, characterized in that, The test conditions include calibration conditions, repeatability conditions, temperature and pressure adaptability conditions, blind zone conditions, and fault self-test conditions.

3. The functional testing method for a sound velocity sensor according to claim 1, characterized in that, The cross-tree connected catalytic heterogeneous decision forest architecture includes: Five heterogeneous decision trees, each corresponding to a test condition. Each heterogeneous decision tree consists of a root node, intermediate nodes, and leaf nodes arranged in a vertical hierarchy. The signal catalytic amplification mechanism and the data traceability storage unit are embedded in the nodes of each heterogeneous decision tree. The root node is used to receive the test signal for the corresponding working condition and distribute the test signal to the intermediate nodes; The intermediate node is used to receive the test signal distributed by the root node. After performing amplitude compensation and interference suppression processing on the test signal under the action of the signal catalytic amplification mechanism, it performs feature splitting determination, and stores the processing result and determination result in the data traceability storage unit before distributing them to the leaf node. The leaf node is used to receive the processing results and judgment results distributed by the intermediate node, store the complete processing link data corresponding to the processing results and judgment results into the data traceability storage unit as the historical traceability data of the test under this working condition, and output the processing results and judgment results as the single working condition judgment results of the heterogeneous decision tree. The heterogeneous decision trees achieve bidirectional data interaction through lateral connecting branches. The lateral connecting branches connect the peer nodes of each heterogeneous decision tree, and are used to retrieve the status data, historical drift feature data and interference feature data of the current node of other heterogeneous decision trees in real time when each heterogeneous decision tree performs signal judgment and error analysis.

4. The functional testing method for a sound velocity sensor according to claim 1, characterized in that, The three-layer test results specifically include: The first layer contains the single-condition judgment results output by each heterogeneous decision tree: the calibration condition heterogeneous decision tree outputs the reference sound velocity correction value and zero-point offset error; the repeatability condition heterogeneous decision tree outputs the measurement stability index and fluctuation standard deviation; the temperature and pressure adaptability condition heterogeneous decision tree outputs the temperature drift compensation residual, pressure drift compensation residual, and environmental interference coefficient; the blind zone condition heterogeneous decision tree outputs the minimum measurable distance, weak signal detection threshold, and signal-to-noise ratio; and the fault self-check condition heterogeneous decision tree outputs the fault classification result, signal distortion level, and abnormal alarm level. The second layer consists of cross-condition coupling interference patterns and cross-tree correlation features mined by each heterogeneous decision tree after interacting with their respective error and drift characteristics in real time through the horizontal connected branches. The cross-condition coupling interference patterns include correlation patterns of calibration offset, repeatability deviation and blind zone expansion caused by temperature changes. The third layer is the global comprehensive health status output after integrating the single-condition judgment results of each heterogeneous decision tree in the first layer with the cross-tree association features in the second layer. The global comprehensive health status includes the comprehensive measurement accuracy level of the sound velocity sensor and the anti-interference robustness score under all operating conditions.

5. A functional testing method for a sound velocity sensor according to claim 3, characterized in that, The signal catalytic amplification mechanism includes: Signal amplitude compensation module and interference suppression module are configured at each node of each heterogeneous decision tree; When the test signal is input to the intermediate node, the signal amplitude compensation module is used to perform amplitude compensation processing on the weak signal in the test signal, and the signal component below the preset detection threshold is increased to the preset recognizable amplitude range. The interference suppression module simultaneously identifies and filters out bubble clutter and electromagnetic clutter in the test signal; The test signal after amplitude compensation and interference suppression is used as the processing result of the current node and as the basis for the current node's judgment.

6. The functional testing method for a sound velocity sensor according to claim 1, characterized in that, The data traceability storage unit is configured at each node of each heterogeneous decision tree and is used to store the following data: The raw test signal data received by the current node each time, the processing result data after processing by the signal catalytic amplification mechanism each time, the feature splitting judgment result data made by the node each time, and the abnormal data identified each time; The data traceability storage unit adds timestamp tags and node location tags to each stored data, which is used to realize the full life cycle traceability of test data in the time dimension and the precise location of the node where abnormal data is located in the spatial dimension.

7. A functional testing method for a sound velocity sensor according to claim 1, characterized in that, Each heterogeneous decision tree interacts with the other heterogeneous decision trees in real time through the lateral connectivity branches to identify cross-condition coupling interference patterns, including: When performing signal judgment and error analysis for the current operating condition, each heterogeneous decision tree sends real-time data requests to the other four heterogeneous decision trees through the horizontal connection branch. Receive current node status data, historical drift feature data, and identified interference feature data returned by other heterogeneous decision trees; The error characteristics of the current node of this heterogeneous decision tree are cross-compared with the returned data of other heterogeneous decision trees to identify abnormal fluctuation patterns that occur simultaneously across multiple operating conditions within the same time window. The abnormal fluctuation patterns include calibration offset, repeatability deterioration and blind zone expansion caused by temperature changes. The identified abnormal fluctuation patterns are defined as cross-operating condition coupling interference laws.

8. A functional testing method for a sound velocity sensor according to claim 1, characterized in that, The isolation of the disturbed abnormal data includes: Data in each node of each heterogeneous decision tree that matches the cross-condition coupling interference pattern is defined as abnormal data. Mark the storage node corresponding to the abnormal data as a failed node; The data stored in the failed node is prevented from participating in the processing and output of subsequent steps; The data stored in each node that is not marked as a failed node is treated as valid data and used in subsequent steps to correct signal processing parameters and judgment thresholds.

9. A functional testing method for a sound velocity sensor according to claim 8, characterized in that, The correction of the signal processing parameters and judgment threshold of the current node includes: The valid data is used as the correction baseline dataset; Extract the signal amplitude characteristics, signal-to-noise ratio characteristics, and error distribution characteristics of each data sample in the corrected benchmark dataset; The signal amplitude compensation coefficient of the current node is adjusted based on the signal amplitude characteristics, the interference suppression strength coefficient of the current node is adjusted based on the signal-to-noise ratio characteristics, and the judgment benchmark value of the current node is adjusted based on the error distribution characteristics. Each heterogeneous decision tree performs feature splitting on each intermediate-processed signal based on the corrected signal amplitude compensation coefficient, interference suppression strength coefficient, and judgment benchmark value, and outputs the single-condition judgment result.

10. A functional testing method for a sound velocity sensor according to claim 3, characterized in that, Extracting historical data and iteratively updating the parameters of the signal amplification mechanism to achieve closed-loop self-optimization of the test architecture includes: Extract the historical raw test signal data, historical processing result data and historical feature splitting judgment result data accumulated in the data traceability storage unit of each leaf node, and arrange them according to the time series to form the historical performance evolution trajectory of each node; Based on the historical performance evolution trajectory, the processing deviation trend of the signal catalytic amplification mechanism of each node under the current parameter configuration is identified. Based on the processing deviation trend, the updated values ​​of the signal amplitude compensation coefficient and interference suppression intensity coefficient of each node are iteratively calculated, and the updated values ​​are used as the input parameters of the signal catalytic amplification mechanism of the corresponding node in the next round of testing; The input parameters updated in each round of iterations are stored in the data traceability storage unit of the corresponding node, forming a complete record chain of parameter iteration updates.

Images