Instrument aging dynamic test method based on bidirectional LSTM and wavelet denoising

Abstract

The application provides an instrument aging dynamic test method based on a bidirectional LSTM and wavelet denoising, comprising the following steps: S1, collecting time series running data sets of the measured instrument; S2, performing wavelet denoising processing on the time series running data sets to obtain a denoised data set; S3, generating a standardized feature matrix; S4, constructing a bidirectional LSTM neural network model; S5, processing the standardized feature matrix by using a sliding window method to generate a training sample set; S6, dividing the training sample set into a training subset and a verification subset according to a ratio of 7:3, and training the bidirectional LSTM neural network model; S7, inputting instrument data collected in real time into the model after processing by S2-S3, and outputting an aging degree prediction value; and S8, comparing the aging degree prediction value output by S7 with a dynamic threshold θ, and triggering a warning signal when the time is triggered. The neural network model is adopted to improve the accuracy and reliability of the test result.

Description

Technical Field

[0001] The present invention relates to the technical field of instrument aging testing, and in particular to an instrument aging dynamic testing method based on bidirectional LSTM and wavelet noise reduction. Background Art

[0002] Aging testing of instruments and meters is a method of accelerating product aging by simulating actual usage environments or applying extreme conditions to detect potential defects and evaluate reliability and stability.

[0003] The main purpose of aging testing is to eliminate prematurely failed products due to process or material defects, ensure the stable performance of instruments in long-term use, and improve product design or production processes through test data.

[0004] Common aging test methods are as follows:

[0005] Environmental aging tests, including high temperature aging, low temperature aging, temperature cycling and damp heat aging;

[0006] Electrical stress aging test, including continuous power-on aging, switching cycle test and overvoltage / overcurrent test;

[0007] Mechanical stress aging test, including vibration test, impact test and long-term operation aging test;

[0008] Comprehensive aging test, combining the above multiple aging tests.

[0009] The above aging test methods usually rely on manual labor and the experience of testers to perform tests. For example, fixed thresholds are used to judge test results, and the test data is processed in a rough manner, resulting in low accuracy and low reliability of the test results. Summary of the Invention

[0010] The present invention provides an instrument aging dynamic testing method based on bidirectional LSTM and wavelet noise reduction, and adopts a neural network model to improve the accuracy and reliability of the test results.

[0011] To achieve the above object, the present invention adopts the following technical solutions:

[0012] The instrument aging dynamic testing method based on bidirectional LSTM and wavelet noise reduction includes:

[0013] S1: Collecting a time series operation data set of the instrument under test, wherein the time series operation data set includes a voltage fluctuation value set, a current offset value set, and a temperature response delay time set;

[0014] S2: Perform wavelet denoising on the voltage fluctuation value set, current offset set and temperature response delay time set respectively to obtain the corresponding denoised data sets;

[0015] S3: For the denoised data set, use the normalization formula x'i = (xi - μi) / σi to generate the normalized feature matrix X = [x1, x2, x3], where x'i is the data before normalization, xi is the data after normalization, μi and σi are the statistical mean and standard deviation of the historical data of the same model instrument, respectively, and x1, x2, and x3 are the voltage characteristics, current characteristics, and temperature response characteristics of the denoised data set after normalization corresponding to the voltage fluctuation value set, current offset set, and temperature response delay time set, respectively;

[0016] S4: Build a bidirectional LSTM neural network model. The number of nodes in the model input layer is set to 3, which is consistent with the dimension of the standardized feature matrix X.

[0017] S5: Use the sliding window method to process the standardized feature matrix X and generate a training sample set;

[0018] S6: Split the training sample set into a training subset and a validation subset with a ratio of 7:3. Use the Adam optimizer to train the bidirectional LSTM neural network model constructed in S4 with the MAE+MSE compound loss function. Terminate training when the validation set loss does not decrease for 50 consecutive epochs.

[0019] S7: The real-time collected instrument data is processed by S2-S3 and then input into the model trained by S6 to output the aging prediction value.

[0020] S8: Compare the aging prediction value output by S7 With dynamic threshold θ, when The warning signal is triggered when θ=θbase+Δθ, θbase comes from the mean time between failures parameter in the instrument technical manual, Δθ is the threshold adjustment amount, is the predicted mean value in the last 30 days, and Ytrueavg is the arithmetic mean of the true value of aging actually detected in the last 30 days.

[0021] In this manual, the instrument aging dynamic testing method based on bidirectional LSTM and wavelet noise reduction also includes:

[0022] S9: Output the S7 continuously The sequence is trend analyzed to generate a visual report containing the aging prediction curve and inflection point markers. The inflection points are identified by the second-order derivative mutation detection algorithm.

[0023] In this manual, the instrument aging dynamic testing method based on bidirectional LSTM and wavelet noise reduction also includes:

[0024] S10: Based on S7 The value dynamically updates the test cycle, and the new cycle calculation formula is Where T new is the updated test period, T old is the initial value of the current test cycle, and k is the attenuation coefficient.

[0025] In this specification, the wavelet noise reduction processing in S2 includes:

[0026] The voltage fluctuation value set is decomposed into 5 layers using dB4 wavelet basis and denoised using sqtwolog threshold rule;

[0027] The current offset set is decomposed into 6 layers using the sym8 wavelet basis, and an adaptive threshold λ = σ√(2lnN) is applied, where σ is the noise standard deviation and N is the signal length;

[0028] The temperature response delay time set is first subjected to median filtering with a window width of 5 points, and then the Haar wavelet is used to perform three-layer decomposition and reconstruction.

[0029] In this manual, the S4 bidirectional LSTM neural network model structure includes:

[0030] First hidden layer: 64-unit bidirectional LSTM, returning the complete time series;

[0031] Second hidden layer: 32-unit bidirectional LSTM, outputting only the final time step state;

[0032] Attention mechanism layer, calculates the time step weight αt = softmax(Wa ht + ba); softmax is the normalized exponential function, ht is the output of the first hidden layer, Wa is the trainable weight matrix, and ba is the bias term;

[0033] The output layer uses the Sigmoid activation function to map the predicted value to the [0,1] interval

[0034] In this specification, in S5, the sliding window method is used to process the standardized feature matrix X, with a window length of T = 24h and a sliding step size of ΔT = 1h, to generate a training sample set D = {(Xi, Yi)|i = 1, ..., N}, where Xi is the time series data intercepted by the i-th sliding window in the standardized feature matrix X, Yi = α·x1i+β·x2i+γ·x3i, and the weight coefficients α, β, and γ are extracted from the historical fault data through the principal component analysis method.

[0035] In this specification, the second-order derivative mutation detection algorithm is:

[0036] Second-order derivative Δt=1min;

[0037] When three consecutive sampling points meet The time is marked as the turning point.

[0038] In this specification, the test cycle adjustment strategy in S10 includes:

[0039] T new ∈[0.3T old ,1.5T old ];

[0040] when When the value is ≥0.7 for 5 consecutive times, T is set new =0.3T old .

[0041] In this manual, S6 model training includes:

[0042] Loss function L = 0.7MAE + 0.3MSE,

[0043] The Adam optimizer parameters β1 = 0.9, β2 = 0.999, and the initial learning rate η = 0.001; the learning rate is reduced to 1 / 10 of the original value when the validation loss stagnates.

[0044] In this specification, S1 specifically includes:

[0045] Place the instrument under test in a temperature control box and execute the temperature cycle program:

[0046] The starting temperature was -20°C and the temperature was increased to +60°C at a rate of 2°C / min;

[0047] Maintain high temperature for 30 minutes;

[0048] Cool down to -20℃ at a rate of 1.5℃ / min;

[0049] Synchronous acquisition:

[0050] Voltage fluctuation value sampling rate ≥ 1kHz;

[0051] Current offset sampling interval ≤ 0.1s;

[0052] Temperature response delay time τ(t) = t2 - t1, where t1 is the time it takes for the temperature control box set value to change, and t2 is the time it takes for the instrument display value to reach ±1% of the target value. The timestamp alignment accuracy is ≤1ms.

[0053] In summary, the present invention has at least the following beneficial effects:

[0054] The present invention adopts a neural network model to improve the accuracy and reliability of test results, and realizes high-precision, adaptive aging state evaluation and closed-loop control. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the technical solutions of 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 paying any creative work.

[0056] Figure 1 This is a schematic diagram of the instrument aging dynamic testing method based on bidirectional LSTM and wavelet denoising involved in the present invention.

[0057] Figure 2 Schematic diagram of the wavelet denoising process in S2 involved in the present invention.

[0058] Figure 3 Schematic diagram of the noise reduction of V(t) involved in the present invention.

[0059] Figure 4 Schematic diagram of noise reduction of I(t) involved in the present invention.

[0060] Figure 5 Schematic diagram of denoising τ(t) involved in the present invention. DETAILED DESCRIPTION

[0061] Hereinafter, only certain exemplary embodiments are briefly described. As will be appreciated by those skilled in the art, the described embodiments may be modified in various ways without departing from the spirit or scope of the embodiments of the present invention. Therefore, the drawings and description are to be regarded as illustrative in nature and not restrictive.

[0062] The disclosure below provides many different embodiments or examples for implementing different structures of the embodiments of the present invention. In order to simplify the disclosure of the embodiments of the present invention, the components and configurations of specific examples are described below. Of course, these are merely examples and are not intended to limit the embodiments of the present invention. In addition, the embodiments of the present invention may repeat reference numerals and / or reference letters in different examples. Such repetition is for the purpose of simplicity and clarity and does not in itself indicate the relationship between the various embodiments and / or configurations discussed.

[0063] The embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0064] like Figure 1 As shown, this embodiment provides a dynamic instrument aging test method based on bidirectional LSTM and wavelet noise reduction, including:

[0065] Step S1: Collecting a time series operation data set of the instrument under test under preset environmental conditions, wherein the time series operation data set includes a voltage fluctuation value set {V(t)}, a current offset set {I(t)}, and a temperature response delay time set {τ(t)}, where t represents a sampling time point;

[0066] Step S2: performing wavelet denoising on {V(t)}, {I(t)}, and {τ(t)} obtained in step S1 to obtain the denoised data sets {V'(t)}, {I'(t)}, and {τ'(t)};

[0067] Step S3: Input the denoised data set into the standardization processing module and generate the standardized feature matrix X = [x1, x2, x3] according to the formula x'i = (xi - μi) / σi, where x1 corresponds to V'(t), x2 corresponds to I'(t), x3 corresponds to τ'(t), and μi and σi are the statistical mean and standard deviation of the historical data of the same model instrument;

[0068] Step S4: construct a bidirectional LSTM neural network model, where the number of nodes in the model input layer is set to 3, which is strictly consistent with the dimension of the standardized feature matrix X;

[0069] Step S5: Use the sliding window method to process the standardized feature matrix X, with a window length of T = 24h and a sliding step size of ΔT = 1h, to generate a training sample set D = {(Xi, Yi)|i = 1, ..., N}, where Yi = α·x1i + β·x2i + γ·x3i, and the weight coefficients α, β, and γ are extracted from the historical fault data using the principal component analysis method;

[0070] Step S6: Split the training sample set D into a training subset Dtrain and a validation subset Dval in a ratio of 7:3. Use the Adam optimizer to train the neural network constructed in step S4 with the MAE+MSE composite loss function. Terminate the training when the validation set loss does not decrease for 50 consecutive epochs.

[0071] Step S7: The real-time collected instrument data is processed in steps S2-S3 and input into the trained model to output the aging prediction value.

[0072] Step S8: Compare the output of step S7 With dynamic threshold θ, when The early warning signal is triggered when θ=θbase+Δθ, where θbase comes from the mean time between failures parameter in the instrument technical manual, and Δθ is dynamically adjusted according to the prediction error of the last 30 days through the online learning module;

[0073] Step S9: The output of step S7 is continuously The sequence is input into a trend analysis module (existing modules such as Python libraries, R language, ARIMA / SARIMA, Holt-Winters, etc.) to generate a visual report containing an aging prediction curve and inflection point markers. The inflection points are identified using a second-order derivative mutation detection algorithm.

[0074] Step S10: According to step S7 The value dynamically updates the test cycle, and the new cycle calculation formula is Where T old is the initial value of the current test cycle, k = 0.85 is the attenuation coefficient, which can be determined by Monte Carlo simulation optimization.

[0075] In some embodiments, the tool for visualizing reports selects:

[0076] Python Ecosystem:

[0077] Matplotlib / Seaborn: static charts (line charts, bar charts);

[0078] Plotly / Dash: interactive dynamic reports (supports hovering, zooming, and data export);

[0079] Bokeh: Visualize real-time data streams.

[0080] Front-end framework:

[0081] ECharts (JavaScript): suitable for web embedding;

[0082] Power BI / Tableau: Enterprise-level automated reporting.

[0083] In some embodiments, the Monte Carlo simulation optimization determines the following:

[0084] 1. Define optimization objectives and constraints

[0085] Optimization goal: Determine the optimal value of the attenuation coefficient k so that the test cycle adjustment strategy maximizes test efficiency (i.e., shortens unnecessary test cycles) while ensuring that the warning missed rate is less than 5%.

[0086] Constraints:

[0087] False Negative Rate (FNR) <5%;

[0088] Test cycle adjustment range limit: T new ∈[0.3T old ,1.5T old ].

[0089] 2. Monte Carlo simulation steps

[0090] Step 1: Parameter space definition

[0091] Candidate range of attenuation coefficient k: k∈[0.5,1.2] (uniform distribution, i.e., kcandidate~U(0.5,1.2)).

[0092] Number of simulations: At least 10,000 random samples to ensure statistical significance.

[0093] Step 2: Aging scenario simulation

[0094] Generate multiple sets of simulated aging data, covering different aging modes:

[0095] Rapid aging: aging degree In the short term (e.g. 10 days), it rises from 0.3 to 0.9;

[0096] Linear aging: Increase at a constant rate (e.g. +0.02 per day);

[0097] Random fluctuation aging: Random fluctuations within the normal range (0.2-0.6);

[0098] Sudden aging: After the stable phase, it suddenly jumps (e.g. from 0.4 to 0.8). Step 3: Simulation run (single simulation process)

[0099] For each kcandidate, do the following:

[0100] Initialization parameters:

[0101] Initial test period T old =24h;

[0102] Aging series Generate according to preset scenarios.

[0103] Cycle-adjusted simulation:

[0104] Calculate T at each time point according to the formula new :

[0105]

[0106] Constraints: If T new Exceeding [0.3T old ,1.5T old ], then take the boundary value.

[0107] False negative detection:

[0108] Record the number of missed aging events due to cycle extension;

[0109] Calculate the false negative rate: FNR = total number of actual aging events × number of missed events × 100%;

[0110] Step 4: Performance Evaluation Metrics

[0111] For each k_candidate, the following metrics are calculated:

[0112] False negative rate (FNR): FNR < 5% must be met;

[0113] Mean Test Interval (MTI):

[0114]

[0115] Resource Utilization (RU):

[0116]

[0117] Step 5: Screen the optimal k value

[0118] Filtering constraint: remove all k_candidates with FNR ≥ 5%;

[0119] Multi-objective optimization: Among the remaining candidate values, select k that satisfies all of the following conditions:

[0120] Minimum MTI (high testing frequency, low risk of missed detection);

[0121] Lowest RU (low resource consumption).

[0122] Robustness verification: Cross-validate the optimal k and repeat the simulation using independent data sets to ensure stability.

[0123] In some embodiments, the process of extracting weight coefficients α, β, and γ by principal component analysis (PCA) is as follows:

[0124] 1. Data preparation and standardization;

[0125] Historical fault data set: collect fault data of the same type of instrument and construct the data matrix Z∈R m×3 , where: m is the number of fault samples;

[0126] Each row represents a fault sample, which contains three features:

[0127]

[0128] x1: voltage fluctuation value, x2: current offset, x3: temperature response delay time.

[0129] Data standardization:

[0130] Normalize each feature column to eliminate dimension differences:

[0131]

[0132] where μj and σ j are the mean and standard deviation of the j-th feature respectively.

[0133] 2. Covariance matrix calculation;

[0134] Calculate the covariance matrix Σ∈R of the standardized data 3×3 :

[0135]

[0136] The covariance matrix reflects the linear correlation between the three features.

[0137] 3. Eigen Decomposition;

[0138] Perform eigendecomposition on the covariance matrix Σ: Σ=UΛU T ;

[0139] Where: U∈R 3×3 is an eigenvector matrix, each column is an eigenvector;

[0140] Λ=diag(λ1,λ2,λ3) is the eigenvalue diagonal matrix, λ1≥λ2≥λ3≥0.

[0141] The first principal component direction: the eigenvector u1 corresponding to the maximum eigenvalue λ1 = [u 11 ,u 12 ,u 13 ] T .

[0142] 4. Extract weight coefficients α, β, and γ;

[0143] Direct assignment method:

[0144] Normalize the eigenvector of the first principal component as the weight coefficient:

[0145]

[0146] Ensure α 2 +β 2 +γ 2 =1.

[0147] Variance contribution adjustment method (optional):

[0148] If the weights need to reflect the actual explanatory power of the principal components, they can be adjusted according to the eigenvalue ratio:

[0149]

[0150] This method makes the weight coefficient proportional to the variance contribution of the principal component.

[0151] In some embodiments, step S1 specifically includes:

[0152] Preset environmental conditions:

[0153] Connect the instrument under test (digital multimeter 34401A) to the standard source generator (FLUKE 5522A);

[0154] The temperature control box performs a temperature cycle program: starting temperature θstart = -20°C, heating rate r = 2°C / min to θmax = +60°C, holding for t = 30min, then cooling back to θstart at r = 1.5°C / min;

[0155] Synchronous acquisition:

[0156] V(t) was sampled at f = 1 kHz using a NIPXIe-4300 data acquisition card;

[0157] I(t) was sampled using the splitter method with Δt = 0.1 s;

[0158] τ(t)=t2-t1, where t1 is the time it takes for the temperature control box set value to change, t2 is the time it takes for the instrument display value to reach ±1% of the target value, and the timestamp alignment accuracy is ≤1ms.

[0159] In some embodiments, the wavelet noise reduction processing in step S2 includes:

[0160] For V(t): use dB4 wavelet basis function to perform L=5-layer decomposition, and the threshold rule is sqtwolog(λ=σ√(2lnN));

[0161] For I(t): use the sym8 wavelet basis function for L=6-layer decomposition, adaptive threshold λ=σ√(2lnN), σ=median(|D1|) / 0.6745;

[0162] For τ(t): first perform median filtering with a window width of W = 5, and then use Haar wavelet to perform L = 3 layer decomposition and reconstruction.

[0163] In some embodiments, in step S3:

[0164] μi and σi are derived from the factory inspection database of the same type of instrument, with a data volume of ≥100 instruments (each instrument contains at least 100 hours of data);

[0165] Update the parameters every time m=50 new data are added:

[0166]

[0167] Where n is the amount of original data, and m is the amount of new data.

[0168] In some embodiments, the bidirectional LSTM neural network model of step S4 includes:

[0169] Input layer: 3 nodes receiving

[0170] First hidden layer: 64-unit bidirectional LSTM (return sequence);

[0171] Second hidden layer: 32-unit bidirectional LSTM (returns the final state);

[0172] Attention layer: weight αt = softmax(Wa ht + ba), where is the output of the first hidden layer;

[0173] Output layer: Sigmoid activation function output

[0174] In some embodiments, in step S5:

[0175] Sliding window parameters: T = 24h corresponds to N = 1440 sampling points (Δt = 1min), ΔT = 1h corresponds to ΔN = 60 points;

[0176] The weight coefficients α=0.5, β=0.3, γ=0.2 were used to analyze the covariance matrix by principal component analysis. Perform eigendecomposition to obtain .

[0177] In some embodiments, the model training in step S6 includes:

[0178] Loss function L = 0.7MAE + 0.3MSE,

[0179] Adam optimizer parameters β1 = 0.9, β2 = 0.999, initial learning rate η = 0.001;

[0180] After training is completed, the model file is stored using AES-256 encryption.

[0181] In some embodiments, the calculation rule of the dynamic threshold θ in step S8 is:

[0182] θbase=1-exp(-Tobs / MTBF), Tobs=72h, MTBF=5000h;

[0183] The forecast mean for the last 30 days;

[0184] θ∈[0.6,0.9], taking the boundary value when it exceeds.

[0185] In some embodiments, the inflection point detection algorithm of step S9 is:

[0186] Second-order derivative Δt=1min;

[0187] When three consecutive sampling points meet The time is marked as the turning point.

[0188] In some embodiments, the test cycle adjustment strategy of step S10 includes:

[0189] T new ∈[0.3T old ,1.5T old ];

[0190] when When the value is ≥0.7 for 5 consecutive times, T is set new =0.3T old ;

[0191] Send the cycle parameters to the temperature control box register address 0003H via the Modbus TCP protocol.

[0192] In a specific embodiment:

[0193] Step S1: Time series operation data collection;

[0194] Connect the instrument under test (such as digital multimeter 34401A) to the standard source generator (model FLUKE 5522A);

[0195] Place the instrument under test in a temperature control box (such as ESPEC PH-032) and set the temperature cycle program:

[0196] Starting temperature: -20℃;

[0197] Heat to +60°C (rate 2°C / min);

[0198] Keep high temperature for 30 minutes;

[0199] Cool down to -20℃ (rate 1.5℃ / min);

[0200] Synchronously start the data acquisition device:

[0201] Voltage fluctuation value V(t): recorded at a sampling rate of 1 kHz using a high-precision data acquisition card (NIPXIe-4300) or a high-precision voltage sensor (such as Keysight 34470A);

[0202] Current offset I(t): measured using the shunt method or a Hall current sensor (such as LEM LAH 50-P), with a sampling interval of 0.1s;

[0203] Temperature response delay τ(t): Record the time t1 between the temperature control box set value change and the time t2 when the instrument display value reaches the target value ±1% range, and calculate τ(t) = t2 - t1; or record the temperature change curve using a thermocouple (such as K type) and calculate τ(t) = t2 - t1;

[0204] Generate the original dataset:

[0205] The storage format is time series {V(t), I(t), τ(t)}, and the timestamp alignment accuracy is ≤1ms;

[0206] Step S2: wavelet noise reduction processing;

[0207] Denoise V(t):

[0208] a. Use dB4 wavelet basis function to perform 5-layer decomposition;

[0209] b. Use the default threshold rule ('sqtwolog') to remove high-frequency noise;

[0210] c. Reconstruct the signal to obtain V'(t);

[0211] Denoise I(t):

[0212] a. Use sym8 wavelet basis function to perform 6-layer decomposition;

[0213] b. Apply an adaptive threshold: λ = σ√(2lnN), where σ is the noise standard deviation and N is the signal length;

[0214] c. Reconstruct the signal to obtain I'(t);

[0215] Denoise τ(t):

[0216] a. First perform median filtering (window width = 5 sampling points);

[0217] b. Perform Haar wavelet 3-layer decomposition and reconstruction;

[0218] c. Output τ'(t);

[0219] Output denoised dataset: {V'(t), I'(t), τ'(t)};

[0220] Step S3: standardization processing;

[0221] Get historical data statistics parameters of the same model instrument:

[0222] mean μv,μi,μτ;

[0223] Standard deviation σv, σi, στ;

[0224] (Source: Instrument factory inspection database, containing data of at least 100 instruments of the same model);

[0225] Perform normalization on the denoised data:

[0226] x1=(V'(t)-μv) / σv;

[0227] x2=(I'(t)-μi) / σi;

[0228] x3=(τ'(t)-μτ) / στ;

[0229] Generate a standardized feature matrix X = [x1, x2, x3] with a dimension of N × 3 (N is the total number of sampling points); Step S4: Neural network construction;

[0230] Build a bidirectional LSTM network structure:

[0231] Input layer: 3 nodes, receiving the standardized feature matrix X;

[0232] Hidden layer 1: Bidirectional LSTM (64 units, return sequence = True);

[0233] Second hidden layer: bidirectional LSTM (32 units);

[0234] Attention layer: calculates the time step weight, the formula is a_t = softmax(Wa ht + ba);

[0235] Output layer: fully connected layer (1 node, Sigmoid activation);

[0236] Parameter initialization:

[0237] Weight: initialized with He normal distribution;

[0238] Bias: initialized to 0.01;

[0239] Step S5: training sample generation;

[0240] Sliding window parameter settings:

[0241] Window length T = 24h (corresponding to 1440 sampling points, interval 1min);

[0242] Sliding step ΔT = 1h (60 sampling points);

[0243] Perform window segmentation on the normalized matrix X:

[0244] Generate input samples Xi = X[i:i+T,:], i = 0, 60, 120, ...

[0245] Compute the supervised label Yi:

[0246] Obtain principal component analysis weight coefficients (α = 0.5, β = 0.3, γ = 0.2);

[0247] Yi = 0.5x1 + 0.3x2 + 0.2x3 (calculated for the last time step in the window);

[0248] Generate a training sample set D = {(X1, Y1), ..., (XN, YN)};

[0249] Step S6: model training;

[0250] Data partitioning:

[0251] Training set Dtrain: the first 70% samples;

[0252] Validation set Dval: last 30% samples;

[0253] Training configuration:

[0254] Optimizer: Adam (β1=0.9,β2=0.999);

[0255] Loss function: L = 0.7MAE + 0.3MSE;

[0256] Batch size: 32;

[0257] Training process:

[0258] Maximum number of iterations: 500 epochs;

[0259] Early stopping condition: the validation loss does not decrease for 50 consecutive epochs;

[0260] Learning rate scheduling: when the loss stagnates, it is reduced to 1 / 10 of the original value;

[0261] Output the model parameter file after training;

[0262] Step S7: real-time prediction;

[0263] Data stream processing:

[0264] Create a ring buffer on the edge device (NVIDIA Jetson Nano);

[0265] Continuously receiving the standardized data stream outputted by step S3;

[0266] When the buffer is full of 24 hours of data, the prediction is triggered;

[0267] Model Inference:

[0268] Load the trained TensorRT engine;

[0269] The input data dimension is adjusted to (1,1440,3);

[0270] Output predicted value (Keep 4 decimal places);

[0271] Store prediction results in a time series database (InfluxDB);

[0272] Step S8: aging warning;

[0273] Dynamic threshold calculation:

[0274] Base threshold θ_base = 0.75 (converted from the instrument manual MTBF = 5000h);

[0275] Adjustment amount (data from the last 30 days);

[0276] Final threshold θ = θbase + Δθ (limited between 0.6 and 0.9);

[0277] Early warning logic:

[0278] when When the warning signal is generated, the three-level warning signal is generated:

[0279] Yellow warning, record log;

[0280] Orange warning, triggering the self-check procedure;

[0281] Red alert, forced shutdown;

[0282] Step S9: Visual report generation;

[0283] Data input:

[0284] Get 72 consecutive hours from the time series database sequence;

[0285] Inflection point detection:

[0286] Compute the second-order derivative:

[0287] Mark the turning point: And it lasts for more than 3 sampling points;

[0288] Generate report content:

[0289] Line chart display Curves changing over time;

[0290] The red color shows the period when the theta value is exceeded;

[0291] The table counts the number of warnings and the highest aging degree;

[0292] Step S10: test cycle adjustment;

[0293] Update formula execution:

[0294] Current cycle T old Read from system configuration files;

[0295] Calculate the new period:

[0296] Constraints: T new ∈[0.3T old ,1.5T old ];

[0297] Update strategy:

[0298] when If the value is ≥0.7 for 5 times in a row, set T immediately. new =0.3T old ;

[0299] In other cases, a periodic update is performed every 24 hours;

[0300] Feedback control:

[0301] T new Write to the configuration register of the test controller;

[0302] Send the "cycle reset" command to the temperature control box via the Modbus TCP protocol.

[0303] The above embodiments are intended to illustrate the present invention, not to limit the present invention. Therefore, changes in illustrative values ​​or substitutions of equivalent components should still fall within the scope of the present invention.

[0304] From the above detailed description, it will be clear to those skilled in the art that the present invention can indeed achieve the aforementioned objectives and is in compliance with the provisions of the Patent Law.

[0305] Although preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they become aware of the basic inventive concepts. Therefore, the appended claims are intended to be interpreted as covering the preferred embodiments and all changes and modifications that fall within the scope of the invention. The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. It should be noted that any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention are intended to be included within the scope of protection of the present invention.

[0306] It should be noted that the above description of the relevant processes is for illustration and purpose only and does not limit the scope of application of this specification. For those skilled in the art, various modifications and changes can be made to the processes under the guidance of this specification. However, such modifications and changes are still within the scope of this specification.

[0307] The basic concepts have been described above. It will be apparent to those skilled in the art after reading this application that the above disclosures are merely illustrative and do not constitute limitations on this application. Although not explicitly stated herein, those skilled in the art may make various modifications, improvements, and amendments to this application. Such modifications, improvements, and amendments are suggested in this application and remain within the spirit and scope of the exemplary embodiments of this application.

[0308] At the same time, this application uses specific terms to describe the embodiments of this application. For example, "one embodiment," "an embodiment," and / or "some embodiments" refer to a certain feature, structure, or characteristic related to at least one embodiment of this application. Therefore, it should be emphasized and noted that "one embodiment," "an embodiment," or "an alternative embodiment" mentioned twice or more in different places in this specification does not necessarily refer to the same embodiment. In addition, certain features, structures, or characteristics in one or more embodiments of this application may be appropriately combined.

[0309] In addition, it will be understood by those skilled in the art that various aspects of the present application can be illustrated and described by a number of patentable categories or situations, including any new and useful combination of processes, machines, products or substances, or any new and useful improvements thereto. Therefore, various aspects of the present application can be implemented entirely by hardware, entirely by software (including firmware, resident software, microcode, etc.), or by a combination of hardware and software. The above hardware or software can all be referred to as "units", "modules" or "systems". In addition, various aspects of the present application can take the form of a computer program product embodied in one or more computer-readable media, wherein computer-readable program code is contained therein.

[0310] The computer program code required for the operation of each part of the application can be written in any one or more programming languages, including object-oriented programming languages ​​such as Java, Scala, Smalltalk, Eiffel, JADE, Emerald, C++, C#, VB.NET, Python, conventional procedural programming languages ​​such as C programming language, VisualBasic, Fortran2103, Perl, COBOL2102, PHP, ABAP, dynamic programming languages ​​such as Python, Ruby and Groovy or other programming languages. The program code can be run completely on the user's computer, or run on the user's computer as an independent software package, or run partly on the user's computer and partly on a remote computer, or run completely on a remote computer or server. In the latter case, the remote computer can be connected to the user's computer through any network form, such as a local area network (LAN) or a wide area network (WAN), or be connected to an external computer (such as by the Internet), or in a cloud computing environment, or used as a service such as software as a service (SaaS).

[0311] In addition, unless expressly stated in the claims, the order of the processing elements and sequences described in this application, the use of alphanumeric characters, or the use of other names are not intended to limit the order of the processes and methods of this application. Although the above disclosure discusses some embodiments of the invention that are currently considered useful through various examples, it should be understood that such details are only for illustrative purposes, and the attached claims are not limited to the disclosed embodiments. On the contrary, the claims are intended to cover all modifications and equivalent combinations that are consistent with the essence and scope of the embodiments of this application. For example, although the implementation of the various components described above can be embodied in a hardware device, it can also be implemented as a pure software solution, for example, installation on an existing server or mobile device.

[0312] Similarly, it should be noted that in order to simplify the presentation of this disclosure and thereby facilitate understanding of one or more of the invention's embodiments, the foregoing descriptions of the embodiments of this disclosure sometimes combine multiple features into a single embodiment, figure, or description thereof. However, this approach should not be interpreted as reflecting an intention that the claimed subject matter requires more features than expressly recited in each claim. Rather, the subject matter of the invention may possess fewer features than the single embodiment described above.

Claims

1. The instrument aging dynamic test method based on bidirectional LSTM and wavelet noise reduction is characterized by: include: S1: Collecting a time series operation data set of the instrument under test, wherein the time series operation data set includes a voltage fluctuation value set, a current offset value set, and a temperature response delay time set; S2: Perform wavelet denoising on the voltage fluctuation value set, current offset set and temperature response delay time set respectively to obtain the corresponding denoised data sets; S3: For the denoised data set, use the normalization formula x'i = (xi - μi) / σi to generate the normalized feature matrix X = [x1, x2, x3], where x'i is the data before normalization, xi is the data after normalization, μi and σi are the statistical mean and standard deviation of the historical data of the same model instrument, respectively, and x1, x2, and x3 are the voltage characteristics, current characteristics, and temperature response characteristics of the denoised data set after normalization corresponding to the voltage fluctuation value set, current offset set, and temperature response delay time set, respectively; S4: Build a bidirectional LSTM neural network model. The number of nodes in the model input layer is set to 3, which is consistent with the dimension of the standardized feature matrix X. S5: Use the sliding window method to process the standardized feature matrix X and generate a training sample set; S6: Split the training sample set into a training subset and a validation subset with a ratio of 7:

3. Use the Adam optimizer to train the bidirectional LSTM neural network model constructed in S4 with the MAE+MSE compound loss function. Terminate training when the validation set loss does not decrease for 50 consecutive epochs. S7: The real-time collected instrument data is processed by S2-S3 and then input into the model trained by S6 to output the aging prediction value. S8: Compare the aging prediction value output by S7 With dynamic threshold θ, when The warning signal is triggered when θ=θbase+Δθ, θbase comes from the mean time between failures parameter in the instrument technical manual, Δθ is the threshold adjustment amount, is the predicted mean value in the last 30 days, and Ytrueavg is the arithmetic mean of the true value of aging actually detected in the last 30 days.

2. The instrument aging dynamic testing method based on bidirectional LSTM and wavelet noise reduction according to claim 1 is characterized in that: Also includes: S9: Output the S7 continuously The sequence is trend analyzed to generate a visual report containing the aging prediction curve and inflection point markers. The inflection points are identified by the second-order derivative mutation detection algorithm.

3. The instrument aging dynamic testing method based on bidirectional LSTM and wavelet noise reduction according to claim 1 is characterized in that: Also includes: S10: Based on S7 The value dynamically updates the test cycle, and the new cycle calculation formula is Where T new is the updated test period, T old is the initial value of the current test cycle, and k is the attenuation coefficient.

4. The instrument aging dynamic testing method based on bidirectional LSTM and wavelet noise reduction according to claim 1 is characterized in that: Wavelet noise reduction processing in S2 includes: The voltage fluctuation value set is decomposed into 5 layers using dB4 wavelet basis and denoised using sqtwolog threshold rule; The current offset set is decomposed into 6 layers using the sym8 wavelet basis and an adaptive threshold is applied. Where σ is the standard deviation of noise, N is the signal length; The temperature response delay time set is first subjected to median filtering with a window width of 5 points, and then the Haar wavelet is used to perform three-layer decomposition and reconstruction.

5. The instrument aging dynamic testing method based on bidirectional LSTM and wavelet noise reduction according to claim 1 is characterized in that: The S4 bidirectional LSTM neural network model structure includes: First hidden layer: 64-unit bidirectional LSTM, returning the complete time series; Second hidden layer: 32-unit bidirectional LSTM, outputting only the final time step state; Attention mechanism layer, calculates the time step weight αt = softmax(Wa ht + ba); softmax is the normalized exponential function, ht is the output of the first hidden layer, Wa is the trainable weight matrix, and ba is the bias term; The output layer uses the Sigmoid activation function to map the predicted value to the [0,1] interval.

6. The instrument aging dynamic testing method based on bidirectional LSTM and wavelet noise reduction according to claim 1 is characterized in that: In S5, the sliding window method is used to process the standardized feature matrix X, with a window length of T = 24h and a sliding step size of ΔT = 1h, to generate a training sample set D = {(Xi, Yi)|i = 1, ..., N}, where Xi is the time series data intercepted by the i-th sliding window in the standardized feature matrix X, Yi = α·x1i+β·x2i+γ·x3i, and the weight coefficients α, β, and γ are extracted from the historical fault data through the principal component analysis method.

7. The instrument aging dynamic testing method based on bidirectional LSTM and wavelet noise reduction according to claim 2 is characterized in that: The second-order derivative mutation detection algorithm is: Second-order derivative Δt=1min; When three consecutive sampling points meet The time is marked as the turning point.

8. The instrument aging dynamic testing method based on bidirectional LSTM and wavelet noise reduction according to claim 3, It is characterized by: The testing cycle adjustment strategies in S10 include: T new ∈[0.3T old ,1.5T old ]; when When the value is ≥0.7 for 5 consecutive times, T is set new =0.3T old .

9. The instrument aging dynamic testing method based on bidirectional LSTM and wavelet noise reduction according to claim 1 is characterized in that: S6 model training includes: Loss function L = 0.7MAE + 0.3MSE, The Adam optimizer parameters β1 = 0.9, β2 = 0.999, the initial learning rate η = 0.001; the learning rate is reduced to 1 / 10 of the original value when the validation loss stagnates, and N is the total number of sampling points.

10. The instrument aging dynamic testing method based on bidirectional LSTM and wavelet noise reduction according to claim 1 is characterized in that: S1 specifically includes: Place the instrument under test in a temperature control box and execute the temperature cycle program: The starting temperature was -20°C and the temperature was increased to +60°C at a rate of 2°C / min; Maintain high temperature for 30 minutes; Cool down to -20℃ at a rate of 1.5℃ / min; Synchronous acquisition: Voltage fluctuation value sampling rate ≥ 1kHz; Current offset sampling interval ≤ 0.1s; Temperature response delay time τ(t) = t2 - t1, where t1 is the time it takes for the temperature control box set value to change, and t2 is the time it takes for the instrument display value to reach ±1% of the target value. The timestamp alignment accuracy is ≤1ms.

Images