Performance test system and equipment of fan grating sensor and storage medium

Abstract

The invention discloses a fan grating sensor performance test system and device and a storage medium, and relates to the technical field of sensor testing, and the system comprises a scene simulation platform building module which is used for building a scene simulation platform; the scene template set acquisition module is used for acquiring a matched scene template set; the grating sensor set recording module is used for traversing the configuration platform and recording a sensor signal sequence; the multi-scale enhanced feature set acquisition module is used for performing time sequence multi-scale analysis to obtain an enhanced feature set; the target multi-scale enhancement feature determination module is used for determining target enhancement features; and the target performance test result determination module is used for analyzing and determining a target performance test result by using a comprehensive analyzer. According to the method, the technical problem of insufficient result reliability caused by incomplete scene coverage and extensive test data processing in a traditional test method is solved, and the technical effects of accurately testing the long-time sequence performance of batch fan grating sensors and improving the result reliability are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of sensor testing, and in particular to a performance testing system, equipment and storage medium for a fan grating sensor. Background Art

[0002] During the operation and maintenance of industrial equipment, the performance and reliability of fan grating sensors are crucial to equipment safety and energy efficiency. In existing technologies, performance testing of fan grating sensors faces the following challenges: Traditional testing methods struggle to simulate diverse real-world failure scenarios (such as light source attenuation, occlusion, and receiver damage). Furthermore, they lack the ability to extract multi-scale features from long-term time series test data, making it impossible to accurately capture subtle trends in sensor performance degradation. For example, when simulating complex operating conditions, traditional methods can only monitor signals in a single dimension, failing to dynamically adapt to failure modes in different application scenarios. Furthermore, their crude processing of multi-scale features such as millisecond-level pulse fluctuations and hour-level performance drift contained in time series data can easily lead to the loss of critical information and unreliable test results. Summary of the Invention

[0003] The present application provides a performance testing system, equipment and storage medium for a fan grating sensor, which are used to solve the technical problems in traditional testing methods such as incomplete scene coverage and rough test data processing resulting in insufficient reliability of results.

[0004] In the first aspect, the present application provides a performance testing system for a fan grating sensor, the system comprising: a scene simulation platform building module for building a scene simulation platform, wherein the scene simulation platform includes an application scenario template library; a scene template set acquisition module for acquiring usage information of a target batch of fan grating sensor sets, matching the application scenario template library based on the usage information, and obtaining a matching application scenario template set; a grating sensor set recording module for traversing the matching application scenario template set to configure the scene simulation platform, and recording the count value-pulse signal amplitude-waveform sequence set of the fan grating sensor set; a multi-scale enhancement feature set acquisition module for performing time series multi-scale analysis on the count value-pulse signal amplitude-waveform sequence set to obtain a time series multi-scale enhancement feature set; a target multi-scale enhancement feature determination module for determining a target time series multi-scale enhancement feature by performing a centralized analysis on the time series multi-scale enhancement feature set; and a target performance test result determination module for analyzing the target time series multi-scale enhancement feature using a performance test comprehensive analyzer to determine a target performance test result.

[0005] In a second aspect, the present application provides an electronic device comprising: a processor; and a memory for storing instructions executable by the processor; wherein the processor is used to execute a performance test system for a fan grating sensor provided in the present application.

[0006] In a third aspect, the present application provides a computer-readable storage medium storing a computer program, which is used to execute a performance testing system for a fan grating sensor provided by the present application.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages: This application builds a scenario simulation platform and configures a multi-dimensional testing process. It obtains sensor performance data through time series multi-scale analysis and feature enhancement processing, and combines the mean shift algorithm and feedforward neural network model to perform feature screening and comprehensive evaluation, thereby realizing long-term performance testing of batch fan grating sensors, solving the problem of low reliability of test results caused by insufficient scene coverage and extensive data processing in traditional methods, making the performance test results of fan grating sensors more accurate and reliable, and achieving the technical effect of accurately testing the long-term performance of batch fan grating sensors and improving the reliability of the results. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0009] Figure 1 A schematic structural diagram of a performance testing system for a fan grating sensor provided in an embodiment of the present application.

[0010] Figure 2 This is a schematic diagram of the structure of the electronic device provided in this application.

[0011] Explanation of the accompanying drawings: scene simulation platform building module 10, scene template set acquisition module 20, grating sensor set recording module 30, multi-scale enhancement feature set acquisition module 40, target multi-scale enhancement feature determination module 50, target performance test result determination module 60, processor 21, memory 22, input device 23, output device 24. DETAILED DESCRIPTION

[0012] This application provides a performance testing system, equipment and storage medium for a fan grating sensor, which is used to solve the technical problems of incomplete scene coverage and rough test data processing in traditional testing methods, resulting in insufficient reliability of results.

[0013] The following will be combined with the accompanying drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making any creative work are within the scope of protection of this application.

[0014] Example 1, as Figure 1 As shown, the present application provides a performance testing system for a fan grating sensor, the system comprising: The scenario simulation platform building module 10 is used to build a scenario simulation platform, wherein the scenario simulation platform includes an application scenario template library.

[0015] In this embodiment, the scenario simulation platform is a component of the performance testing system. Its function is to match application scenario templates and configure test scenarios based on the usage information of the target batch of wind turbine grating sensors. The application scenario template library is a component of the scenario simulation platform, which is composed of multiple application scenario templates.

[0016] Specifically, first, the system uses channels such as sensor deployment archives and user demand forms to obtain usage information of the target batch of sensors. For example, the usage information of a batch of 200 sensors includes key tags such as inland wind power-dust environment (accounting for 60%), offshore wind power-high humidity environment (accounting for 30%), and plateau wind power-low pressure environment (accounting for 10%). Each piece of information is associated with specific operating parameters, such as dust concentration of 50-200mg / m³ and humidity of 85%-95%RH.

[0017] Based on the above usage information, the system starts the application scenario template library matching mechanism, where the specific construction steps of the application scenario template library are detailed in the process from the application scenario mining unit to the application scenario template library acquisition unit. The template library pre-stores standardized scenario templates corresponding to the usage labels, such as templates for inland wind power labels associated with dust particles blocking the light path and temperature differences causing condensation on light source lenses. Each template contains a fault parameter range, such as an obstruction particle size of 0.1-5mm and a condensation duration of 4-12 hours. Through the keyword matching algorithm, the inland wind power-dust environment usage information is accurately matched with the template labeled inland wind power-particulate matter pollution in the template library, and the parameter set of the template is extracted, such as the simulated dust particle size average of 2mm and the concentration average of 150mg / m³, forming a set containing multiple matching templates.

[0018] The system then automatically configures the scenario simulation platform by traversing the set of matching templates. For example, using the sand and dust particle obstruction light path template, a programmable particle generator generates simulated dust particles with a particle size of 2±0.5mm and continuously injects them into the test chamber at a concentration of 150±30mg / m³. Simultaneously, the light intensity is attenuated to 70% of normal operating conditions using the light source adjustment module to simulate the decrease in light transmittance caused by dust adhesion. During the configuration process, the data acquisition module is synchronously triggered to record sensor parameters such as count values, pulse signal amplitude, and waveform sequence at a frequency of 100Hz, forming a multidimensional dataset with a time series dimension.

[0019] By obtaining sensor usage information, matching template library scenarios based on tags, automatically configuring multi-dimensional test conditions, and synchronously collecting timing data, the dynamic construction of the scenario simulation platform is achieved, achieving the effect of accurately reproducing diverse fault scenarios according to actual application needs and efficiently obtaining comprehensive test data.

[0020] The scene template set acquisition module 20 is used to obtain usage information of the target batch of fan grating sensor sets, match the application scene template library based on the usage information, and obtain a matching application scene template set.

[0021] Specifically, first, the system obtains the usage information of the target batch of sensors through the sensor production file interface or manual input interface. For example, the usage information of a batch of 300 sensors includes three core tags: inland wind power-dust conditions (180 units, accounting for 60%), factory workshop-oil pollution environment (90 units, accounting for 30%), and laboratory high-precision monitoring (30 units, accounting for 10%). Each tag is associated with specific parameters, such as the particle concentration range of 50-300mg / m³ in dust conditions and the oil mist particle size of 0.5-2 in oil pollution environments. .

[0022] After obtaining the usage information, the system pre-processes the data: extracting key feature words through natural language processing (NLP) technology, for example, parsing out keywords such as inland wind power, dust, and particulate matter concentration from inland wind power-dust conditions, and converting them into structured data (such as label vector [inland wind power, dust, particulate matter pollution]). At the same time, each template in the application scenario template library has been pre-labeled with standardized labels. For example, the label of the inland wind power-particulate matter occlusion template is [inland wind power, dust, particle size 0.1-5mm, concentration ≤200mg / m³], and the label of the industrial oil pollution-light path pollution template is [factory workshop, oil mist, particle size ≤2 , adhesion rate 0.1-0.5mg / cm² / h].

[0023] Based on the preprocessed usage label vector (for example, extracting keywords such as inland wind power, dust, and particulate matter concentration from the inland wind power-dust condition and converting them into a structured vector) and the template library label vector (for example, the inland wind power, dust, and particle size 0.1-5 mm label vectors from the inland wind power-particulate matter occlusion template), the cosine similarity algorithm calculates similarity through the following steps: First, the two vectors are dimensionally aligned, with identical keywords assigned to the same dimension and missing keywords assigned to zero. The dot product of the two vectors is then calculated, which is the sum of the corresponding dimension values. The modulus of the two vectors is then calculated, which is the square root of the sum of the squares of the values ​​in each dimension. Finally, the dot product is divided by the product of the moduli of the two vectors to obtain a cosine similarity value in the range [-1, 1]. Positive values ​​indicate consistent directions, and larger values ​​indicate higher similarity. For example, the inland wind power-dust condition vector and the inland wind power-particulate matter occlusion template vector share core dimensions such as inland wind power and dust, resulting in high dot product values ​​and similar moduli, leading to high similarity. However, the offshore wind power-salt spray corrosion template vector has a low dot product value and a large difference in modulus, resulting in low similarity, due to significant differences in core dimensions (e.g., inland wind power vs. offshore wind power, dust vs. salt spray). By setting a threshold (e.g., ≥0.8), highly matching templates can be screened out, thereby obtaining a set of matching application scenario templates.

[0024] Through structured extraction of sensor usage information, standardized label preprocessing, intelligent similarity calculation and threshold screening, we have achieved automated and precise mapping from the actual use of sensors to test scenarios, achieving the effect of quickly generating a set of highly compatible test scenarios based on data-driven methods and improving the comprehensiveness and reliability of tests.

[0025] The grating sensor set recording module 30 is used to traverse the matching application scenario template set to configure the scenario simulation platform and record the count value-pulse signal amplitude-waveform sequence set of the fan grating sensor set.

[0026] Specifically, first, the system obtains a set of matching application scenario templates. For example, a batch of sensors matches three templates: light source attenuation, dynamic occlusion, and receiver noise. Each template has a predefined parameter set, such as the light intensity range of the light source attenuation template is 50%-80% and the attenuation rate is 0.5% / min.

[0027] Then, the system starts the automatic traversal configuration according to the template set sequence. Taking the dynamic occlusion template as an example, the programmable mechanical device controls the occlusion to pass through the light path with a particle size of 0.1-5mm and a frequency of 10-20 / second. At the same time, the light source module is adjusted to output a stable light intensity, such as 5000 lux, and the data acquisition module is triggered to synchronously record three core data at a frequency of 1000Hz: count value (range 0-20000 counts, reflecting the occlusion frequency), pulse signal amplitude (0-5V, reflecting the degree of light intensity attenuation), waveform sequence (rising edge / falling edge time, resolution 0.1 , reflecting signal distortion), and then obtaining the count value-pulse signal amplitude-waveform sequence set.

[0028] During the traversal process, the system verifies the accuracy of configuration parameters in real time through the status monitoring module (a functional module used to monitor the accuracy of configuration parameters in real time and ensure consistent reproduction of fault scenarios). For example, the error between the light intensity measured by the optical power meter and the preset template value is ≤±2%. If an anomaly occurs, such as a fluctuation in the obstruction transmission rate exceeding 5%, the system automatically triggers parameter compensation mechanisms, such as adjusting the motor speed, to ensure consistent scene reproduction. For example, in the receiver noise template test, simulated electrical noise is injected with an amplitude of 10-50mV and a frequency of 1-10kHz. The signal-to-noise ratio of the sensor output signal is simultaneously monitored, and the pulse distortion rate under noise interference is recorded. A value of ≤5 distorted pulses per 1000 pulses is considered normal.

[0029] Through automated traversal and matching templates, simultaneous high-density acquisition of multiple parameters, and real-time parameter verification and compensation, the problems of inefficient scene configuration and one-sided data collection in existing technologies have been solved. This allows for rapid reproduction of diverse fault scenarios based on standardized templates and comprehensive acquisition of high-dimensional time series test data, providing a reliable data foundation for subsequent multi-scale feature analysis and performance evaluation.

[0030] The multi-scale enhanced feature set acquisition module 40 is configured to perform a time series multi-scale analysis on the count value-pulse signal amplitude-waveform sequence set to obtain a time series multi-scale enhanced feature set.

[0031] In an embodiment of the present application, time series multi-scale analysis refers to the process of performing multi-dimensional feature extraction and enhancement on the count value-pulse signal amplitude-waveform sequence set recorded by the fan grating sensor in a long time series test through a pre-constructed time series multi-scale convolution channel set.

[0032] Specifically, the system pre-constructs a set of time-series multi-scale convolution channels, extracts and analyzes the first count value-pulse signal amplitude-waveform sequence to obtain a first time-series multi-scale feature set, obtains a first time-series multi-scale enhanced feature through interactive enhancement, and then performs a time-series multi-scale analysis on the overall signal sequence set to obtain a time-series multi-scale enhanced feature set. The specific steps are described in detail from the convolution channel construction unit to the enhanced feature set acquisition unit.

[0033] The target multi-scale enhancement feature determination module 50 is configured to determine the target temporal multi-scale enhancement feature by collectively analyzing the temporal multi-scale enhancement feature set.

[0034] Specifically, the system determines the central feature of the temporal multi-scale enhancement feature set and iteratively updates the central feature using the mean shift algorithm to obtain the target temporal multi-scale enhancement feature. The specific steps are described in detail in the multi-scale enhancement feature determination and acquisition unit.

[0035] The target performance test result determination module 60 uses a performance test comprehensive analyzer to analyze the target temporal multi-scale enhancement features to determine the target performance test result.

[0036] In the embodiment of the present application, the performance test comprehensive analyzer is a model constructed based on a feedforward neural network.

[0037] Specifically, the selected multi-scale enhanced features of the target time series, such as multi-dimensional feature vectors containing short-time pulse distortion rate, medium-time signal attenuation trend, and long-time count drift, are first input into the performance test comprehensive analyzer. This analyzer has a built-in pre-trained feedforward neural network model that automatically extracts nonlinear correlations between features. The specific construction process is detailed in the sample data acquisition unit and the comprehensive analyzer acquisition unit.

[0038] Taking the target features of a batch of sensors as an example, the input vector contains eigenvalues ​​at three scales: the average rising edge time of the waveform at scale 1 (50ms time window) ranges from 0.3 Extended to 0.5 The signal amplitude standard deviation at scale 2 (10-minute time window) increased by 40%, and the count value drift rate at scale 3 (24-hour time window) reached -15%. The analyzer mapped these features into early-stage light source degradation fault categories through multi-layer nonlinear transformations, outputting a 92% confidence level and generating a target performance test result of 75 points (out of a maximum of 100, with scores below 80 triggering a warning).

[0039] By inputting the target time series multi-scale enhanced features into the pre-trained model and utilizing its nonlinear mapping capability to automatically generate performance evaluation results, the sensor performance test results can be quickly and accurately determined based on multi-dimensional features, thereby improving the testing efficiency and decision reliability in industrial scenarios.

[0040] In one possible implementation, the scenario simulation platform building module 10 further includes: An application scenario mining unit is used to mine the application scenarios of the fan grating sensor to obtain a sample application scenario information set and a sample usage information set; an application scenario information acquisition unit is used to perform similar aggregation on the sample usage information set, and map and aggregate the sample application scenario information set according to the aggregation result to obtain multiple aggregated sample application scenario information sets; an application scenario template acquisition unit is used to perform application scenario information mean processing on the multiple aggregated sample application scenario information sets respectively to obtain multiple application scenario templates; an application scenario template library acquisition unit is used to summarize the multiple application scenario templates to obtain the application scenario template library.

[0041] Specifically, first, by comprehensively exploring the actual application scenarios of fan grating sensors, technical personnel in this field rely on historical operation and maintenance data, sensor deployment environment surveys and other channels to collect a large amount of sample application scenario information (such as specific fault types, environmental parameters, working condition duration, etc.) and corresponding sample usage information (such as the type of fan that the sensor is applicable to, installation location, operating cycle, etc.), forming an initial sample data set.

[0042] Subsequently, the sample usage information set is clustered into similar categories. Using the K-means algorithm, each piece of usage information is first converted into a feature vector containing dimensions such as turbine type (e.g., offshore / inland / highland wind power) and installation environment (e.g., humidity / dust / air pressure). For example, offshore wind power in a high-humidity environment is represented as [1, 0, 0, 0.8, 0.2, 0.1] (assuming the first three dimensions are one-hot encodings of the turbine type, and the last three dimensions are normalized values ​​of the environmental parameters). Five to eight cluster centers are then randomly initialized, such as by selecting the feature vectors of a typical scenario as the initial center. Each sample is assigned to the cluster with the closest center by calculating the Euclidean distance. Each cluster center is then iteratively updated to the mean vector of the samples within the cluster until the center converges or the maximum number of iterations (e.g., 50) is reached. Ultimately, five to eight clustered usage categories are formed, such as offshore wind power in a high-humidity environment, inland wind power in a dust-and-air environment, and highland wind power in a low-pressure environment.

[0043] After completing the usage aggregation, the sample application scenario information sets are mapped and aggregated using each usage category as an index. Specifically, for each usage category, all associated application scenario information is extracted to form multiple independent aggregated sample application scenario information sets. For example, within the offshore wind power category, information sets for typical scenarios such as light source corrosion caused by high humidity and salt spray particles blocking the light path can be aggregated. This includes data features such as the frequency of these scenarios in historical data and the range of environmental parameters (e.g., humidity 80%-95%).

[0044] Next, the application scenario information set of each aggregated sample is processed by averaging the application scenario information. By calculating the statistics of each scenario parameter, such as the mean and variance, the core characteristic parameters that can represent this type of scenario are extracted to form a standardized application scenario template. Taking the light source attenuation scenario as an example, the attenuation amplitude (e.g., 20%-50%) and duration (e.g., 4-12 hours) of all light source attenuation events in the aggregated set are averaged to obtain the scene template parameters with an attenuation amplitude mean of 35% and a duration mean of 8 hours. The variance is retained as the parameter fluctuation range, such as ±5% amplitude and ±2 hours duration, thus constructing a statistically representative single scene template.

[0045] The multiple application scenario templates generated by the above process are aggregated to form an application scenario template library. This template library covers typical failure scenarios and their standardized parameters under different usage categories. For example, it contains multiple sub-scenario templates covering multiple failure modes such as light sources, optical paths, and receivers. Each template is labeled with the usage category, such as offshore wind power - high humidity corrosion, inland wind power - sand and dust obstruction, etc., to facilitate subsequent rapid matching and call based on the target sensor's usage information.

[0046] Through the above steps, based on a data-driven approach, a full-process construction from actual application scenario mining to standardized template generation was achieved, enabling the scenario simulation platform to dynamically match a variety of fault scenarios according to the actual use of the sensor, significantly improving the comprehensiveness of performance testing and the accuracy of scenario adaptation.

[0047] In one possible implementation, the multi-scale enhanced feature set acquisition module 40 further includes: A convolution channel construction unit is used to pre-construct a time series multi-scale convolution channel set; a signal sequence acquisition unit is used to extract a first count value-pulse signal amplitude-waveform sequence from the count value-pulse signal amplitude-waveform sequence set; a first feature set acquisition unit uses the time series multi-scale convolution channel set to perform convolution feature analysis on the first count value-pulse signal amplitude-waveform sequence respectively to obtain a first time series multi-scale feature set; a first enhanced feature acquisition unit is used to interactively enhance the first time series multi-scale feature set to obtain a first time series multi-scale enhanced feature; an enhanced feature set acquisition unit uses the time series multi-scale convolution channel set to perform time series multi-scale analysis on the count value-pulse signal amplitude-waveform sequence set to obtain a time series multi-scale enhanced feature set.

[0048] In the embodiment of the present application, the temporal multi-scale convolution channel is a pre-built set of convolution channels with different dilation rates and convolution kernel sizes. Interaction enhancement is the process of strengthening the correlation between features in the temporal multi-scale feature set.

[0049] Specifically, a pre-built set of time-series multi-scale convolution channels with different dilation rates and convolution kernel sizes is first constructed. For example, based on the characteristics of count values, pulse signal amplitudes, and waveform sequences, three sets of convolution channels are constructed: Channel 1, with a convolution kernel size of 3 and a dilation rate of 1, captures short-term time-series details, such as the time variation of a single pulse rise edge; Channel 2, with a convolution kernel size of 9 and a dilation rate of 3, extracts medium-term time-series trends, such as signal amplitude fluctuations within 10 minutes; and Channel 3, with a convolution kernel size of 15 and a dilation rate of 5, analyzes long-term time-series drift, such as the 24-hour count value decay trend. This set optimizes parameters through offline training to cover a time scale range of 0.1ms to 24 hours.

[0050] Next, extract the first count value, pulse signal amplitude, and waveform sequence from the raw data. For example, consider a sensor signal sequence from a light source attenuation scenario, sampled at 100 Hz for one hour, containing 360,000 data points. Using a sliding window technique, split the sequence into multiple subsequences, such as 500 points per window with a 50% overlap, to form a sample set suitable for input into the convolutional network.

[0051] Next, multi-scale convolutional channels are used to perform parallel feature extraction on a single sequence: each channel outputs a corresponding time series feature map, which is then reduced in dimension by a pooling layer to generate a first time series multi-scale feature set. For example, channel 1 outputs 50-dimensional short time series features, such as the pulse width coefficient of variation; channel 2 outputs 30-dimensional medium time series features, such as the amplitude mean shift rate; and channel 3 outputs 20-dimensional long time series features, such as the count value decay slope, ultimately forming a 100-dimensional first time series multi-scale feature set.

[0052] Then, the system randomly extracts features from the first temporal multi-scale feature set without replacement, performs inner product mapping, normalizes the features, constructs an enhancement matrix, and performs convolution enhancement. It then iteratively extracts features for enhancement to obtain the first temporal multi-scale enhanced features. The specific steps are described in detail in the feature similarity set acquisition subunit to the first temporal enhancement feature acquisition subunit.

[0053] Finally, the aforementioned multi-scale convolutional channel and interactive enhancement mechanism are applied to the overall count value, pulse signal amplitude, and waveform sequence. Through batch processing, a time series multi-scale enhanced feature set is generated that encompasses all sensor and full-time data. For example, with 100 sensors and 24 hours of test data, a 100×1440×100-dimensional feature tensor (1440 represents the number of hourly time steps) can be generated, thereby improving data utilization.

[0054] By pre-building a multi-scale convolution channel set, extracting and enhancing single-sequence features, and batch processing the entire data set, we can fully capture multi-dimensional features from long-term time series signals and improve feature expression capabilities, laying a data foundation for the subsequent accurate evaluation of sensor performance.

[0055] In one possible implementation, the first enhanced feature acquisition unit further includes: A feature similarity set acquisition subunit is used to randomly extract two first temporal multi-scale features from the first temporal multi-scale feature set without replacement, perform inner product mapping on the two first temporal multi-scale features, and obtain a feature similarity set; a first enhanced feature acquisition subunit is used to normalize the feature similarity set, construct an enhancement matrix according to the processing result, and use the enhancement matrix to perform convolution enhancement on any one of the two first temporal multi-scale features to obtain a first temporal multi-scale stage enhanced feature; a first temporal enhanced feature acquisition subunit is used to randomly extract a first temporal multi-scale feature from the first temporal multi-scale feature set without replacement, perform feature enhancement on it using the first temporal multi-scale stage enhanced feature, and so on, to obtain the first temporal multi-scale enhanced feature.

[0056] In the embodiment of the present application, inner product mapping is an operation of randomly extracting two features from the first temporal multi-scale feature set without replacement, and mapping them to obtain a feature similarity set by calculating the inner product of the two first temporal multi-scale features.

[0057] Specifically, we first randomly sampled without replacement from the first time series multi-scale feature set (assuming it contains 100-dimensional features), extracting two feature vectors at a time, such as feature A and feature B. We then calculated their similarity using an inner product operation: the inner product value was divided by the product of the two vector moduli to obtain a similarity value in the range [-1, 1]. This sampling was repeated 100 times to form a set of 100 similarity values. For example, the similarity between feature A (the coefficient of variation of the pulse width of a short time series) and feature B (the standard deviation of the amplitude of a medium time series signal) was 0.78, indicating a positive correlation between the two.

[0058] The feature similarity set is then normalized, mapping the value range to [0, 1]. Based on the normalized results, a 5×5 enhancement matrix is ​​constructed, with diagonal elements set to 1 and off-diagonal elements representing the similarity values ​​of corresponding feature pairs. This matrix is ​​used to perform a convolution operation on any feature vector. For example, when enhancing feature A, the convolution of the matrix with feature A increases the weight of the dimension in feature A that is related to feature B, generating the enhanced features of the first temporal multi-scale stage.

[0059] Finally, a single feature, such as feature C, is randomly extracted from the original feature set without replacement, and is successively enhanced using the generated stage-enhanced features (such as the enhanced feature A): the enhanced information of feature A is transferred to feature C through matrix convolution, and this process is repeated until all features are enhanced once, finally forming the first temporal multi-scale enhanced feature.

[0060] Through the steps of random feature pair extraction, similarity calculation, matrix enhancement and iterative transmission, the problem of insufficient feature correlation mining in the existing technology is solved, and the multi-scale feature expression ability is enhanced by feature interaction enhancement, thereby improving the accuracy of sensor performance analysis and improving the accuracy of fault diagnosis.

[0061] In one possible implementation, the target multi-scale enhancement feature determination module 50 further includes: The multi-scale enhancement feature determination unit is used to determine the central temporal multi-scale enhancement feature of the temporal multi-scale enhancement feature set; the multi-scale enhancement feature acquisition unit uses a mean shift algorithm to iteratively update the central temporal multi-scale enhancement feature in the temporal multi-scale enhancement feature set to obtain the target temporal multi-scale enhancement feature.

[0062] Specifically, we first calculate an initial central feature from the temporal multi-scale enhanced feature set. For example, assuming the set contains 1000 feature vectors, each with 50 dimensions, we calculate the mean of each dimension, such as 0.8 for the first dimension and 1.2 for the second dimension, to generate an initial central temporal multi-scale enhanced feature as the starting point for the mean shift algorithm. This initial center represents the average state of the feature set but may not fully reflect the actual dense areas of the data distribution.

[0063] Next, the mean shift algorithm is used to iteratively update the central feature: Taking the initial center as the base point, the weighted mean of all feature vectors within its neighborhood (e.g., within a Euclidean distance range of radius 2) is calculated. The weight is determined by the distance between the feature and the center; the closer the distance, the higher the weight. This results in the new center position. This process is repeated until the change in the center position is less than a preset threshold (e.g., 0.01) or the maximum number of iterations (e.g., 50) is reached. For example, after the first iteration, the center moves 0.5 units toward an area with higher feature density. After 15 iterations, convergence is achieved, and the target temporal multi-scale enhancement feature is finally determined to be located in the peak area of ​​the feature distribution, representing the most representative performance degradation pattern.

[0064] By calculating the initial mean to determine the central feature and using the mean shift algorithm to iteratively optimize the center position, the problems of dispersed feature distribution and lack of prominence of key information in existing technologies are solved, and the effect of accurately extracting highly representative target features from multi-scale feature sets and improving the reliability of performance test results is achieved.

[0065] In one possible implementation, the target performance test result determination module 60 further includes: A sample data acquisition unit is used to obtain multiple sample time series multi-scale enhancement features and multiple sample performance test results as sample data; a comprehensive analyzer acquisition unit is used to divide the sample data into a training set and a verification set according to a preset ratio, and supervise the training of a framework constructed based on a feedforward neural network based on the training set, and use the verification set to verify the trained framework until convergence, thereby obtaining the trained performance test comprehensive analyzer.

[0066] In the embodiment of the present application, the feedforward neural network is a framework for constructing a comprehensive performance test analyzer.

[0067] Specifically, the system first collects multiple sample time series multi-scale enhanced features and corresponding sample performance test results from historical test data as input data. For example, 1,000 sets of data are obtained from long-term time series tests of 100 sensors. Each set of data contains 50-dimensional time series multi-scale enhanced features (such as pulse distortion rate and signal attenuation trends in different time windows) and corresponding performance labels (such as normal, light source attenuation warning, and receiver failure), forming a sample dataset consisting of feature vectors and labels.

[0068] The sample data was then standardized, such as normalizing the feature vectors to [-1, 1], and divided into a training set (800 groups) and a validation set (200 groups) according to a preset ratio (e.g., 8:2). A feedforward neural network framework was trained using the training set: the network input layer received a 50-dimensional feature vector, the hidden layer (assuming two layers, 100 neurons each) extracted nonlinear features using the ReLU activation function, and the output layer used the Softmax function to output the probability distribution of three performance labels. The training process used a cross-entropy loss function and the Adam optimizer. After 500 iterations, the training set loss decreased, while the validation set accuracy increased to .

[0069] During the validation phase, the model was evaluated in real time using a validation set: 200 feature vectors were input, and the model's predicted labels were compared with the true labels. Precision and recall rates exceeding 90% indicated the model had converged to a stable state. The resulting trained performance test analyzer automatically mapped the input time-series multi-scale enhanced features to specific performance results. For example, given a sensor's feature vector, the output was a receiver fault label with a confidence level of 98%.

[0070] Through the above steps, we achieved the effect of training an intelligent analysis model based on large-scale sample data and realizing accurate prediction of fan grating sensor performance, thereby improving the accuracy of cross-scenario fault identification and significantly improving test efficiency and reliability.

[0071] Example 2: Figure 2 This is a structural diagram of an electronic device provided in accordance with a third embodiment of the present invention, showing a block diagram of an exemplary electronic device suitable for implementing the embodiments of the present invention. Figure 2 The electronic device shown is only an example and should not limit the functions and scope of use of the embodiments of the present invention. Figure 2 As shown, the electronic device includes a processor 21, a memory 22, an input device 23 and an output device 24; the number of processors 21 in the electronic device can be one or more. Figure 2 Taking a processor 21 as an example, the processor 21, memory 22, input device 23 and output device 24 in the electronic device can be connected through a bus or other means. Figure 2The bus connection is taken as an example.

[0072] Memory 22, as a computer-readable storage medium, can be used to store software programs, computer-executable programs, and modules, such as the program instructions / modules corresponding to the fire monitoring system incorporating IoT collaborative sensing in the embodiments of this application. Processor 21 executes the software programs, instructions, and modules stored in memory 22 to perform various computer functions and data processing, thereby implementing the aforementioned fan grating sensor performance testing system.

[0073] Any step of the method described above can be stored as a computer instruction or program in an unlimited computer memory 22, and can be called and identified by an unlimited computer processor 21 to implement any method in the embodiments of the present application, without any unnecessary restrictions.

[0074] Furthermore, the terms "first" or "second" as described above not only represent an order relationship but also represent specific concepts and / or refer to the selectability of multiple elements, either individually or in combination. Obviously, those skilled in the art may make various modifications and variations to this application without departing from the scope of this application. Thus, if such modifications and variations fall within the scope of this application and its equivalents, this application is intended to include such modifications and variations.

Claims

1. A performance test system for a fan grating sensor, characterized in that: The system comprises: A scenario simulation platform building module, used to build a scenario simulation platform, wherein the scenario simulation platform includes an application scenario template library; A scene template set acquisition module is used to obtain usage information of a target batch of fan grating sensor sets, and match the application scene template library based on the usage information to obtain a matching application scene template set; A grating sensor set recording module, configured to traverse the matching application scenario template set to configure the scenario simulation platform, and record the count value-pulse signal amplitude-waveform sequence set of the fan grating sensor set; A multi-scale enhanced feature set acquisition module is used to perform a time series multi-scale analysis on the count value-pulse signal amplitude-waveform sequence set to obtain a time series multi-scale enhanced feature set; a target multi-scale enhancement feature determination module, configured to determine a target temporal multi-scale enhancement feature by centrally analyzing the temporal multi-scale enhancement feature set; The target performance test result determination module uses a performance test comprehensive analyzer to analyze the target time series multi-scale enhancement features to determine the target performance test result.

2. A performance test system for a fan grating sensor according to claim 1, characterized in that: include: An application scenario mining unit is used to mine application scenarios of fan grating sensors to obtain a sample application scenario information set and a sample usage information set; an application scenario information acquisition unit, configured to aggregate the sample usage information sets of the same type, and map and aggregate the sample application scenario information sets according to the aggregation results to obtain a plurality of aggregated sample application scenario information sets; an application scenario template acquiring unit, configured to perform application scenario information mean processing on each of the plurality of aggregated sample application scenario information sets to obtain a plurality of application scenario templates; The application scenario template library acquisition unit is configured to aggregate the multiple application scenario templates to obtain the application scenario template library.

3. A performance test system for a fan grating sensor according to claim 1, characterized in that: include: Convolution channel construction unit, used to pre-construct a set of temporal multi-scale convolution channels; a signal sequence acquisition unit, configured to extract a first count value-pulse signal amplitude-waveform sequence from the count value-pulse signal amplitude-waveform sequence set; A first feature set acquisition unit is configured to perform convolution feature analysis on the first count value-pulse signal amplitude-waveform sequence using the time series multi-scale convolution channel set to obtain a first time series multi-scale feature set; A first enhanced feature acquisition unit, configured to interactively enhance the first temporal multi-scale feature set to obtain a first temporal multi-scale enhanced feature; The enhanced feature set acquisition unit uses the time series multi-scale convolution channel set to perform time series multi-scale analysis on the count value-pulse signal amplitude-waveform sequence set to obtain a time series multi-scale enhanced feature set.

4. A performance test system for a fan grating sensor according to claim 3, characterized in that: include: a feature similarity set acquisition subunit, configured to randomly extract two first temporal multi-scale features from the first temporal multi-scale feature set without replacement, perform inner product mapping on the two first temporal multi-scale features, and obtain a feature similarity set; a first enhanced feature acquisition subunit, configured to perform normalization processing on the feature similarity set, construct an enhancement matrix based on the processing result, and use the enhancement matrix to perform convolution enhancement on any one of the two first temporal multi-scale features to obtain a first temporal multi-scale stage enhanced feature; The first temporal enhancement feature acquisition subunit is configured to randomly extract a first temporal multi-scale feature from the first temporal multi-scale feature set without replacement, enhance the feature using the first temporal multi-scale stage enhancement feature, and so on to obtain the first temporal multi-scale enhancement feature.

5. A performance test system for a fan grating sensor according to claim 1, characterized in that: include: a multi-scale enhancement feature determination unit, configured to determine a central temporal multi-scale enhancement feature of the temporal multi-scale enhancement feature set; The multi-scale enhancement feature acquisition unit uses a mean shift algorithm to iteratively update the central temporal multi-scale enhancement feature in the temporal multi-scale enhancement feature set to obtain the target temporal multi-scale enhancement feature.

6. A performance test system for a fan grating sensor according to claim 1, characterized in that: include: A sample data acquisition unit, configured to acquire a plurality of sample time series multi-scale enhancement features and a plurality of sample performance test results as sample data; The comprehensive analyzer acquisition unit is used to divide the sample data into a training set and a verification set according to a preset ratio, perform supervised training on a framework constructed based on a feedforward neural network based on the training set, and verify the trained framework using the verification set until convergence, thereby obtaining the trained performance test comprehensive analyzer.

7. An electronic device, characterized in that: The electronic device comprises: processor; a memory for storing instructions executable by the processor; Wherein, the processor is used to execute a performance test system for a fan grating sensor according to any one of claims 1 to 6.

8. A computer-readable storage medium, characterized in that The storage medium stores a computer program, so the computer program is used to execute the performance test system for a fan grating sensor according to any one of claims 1 to 6.

Images