Portable detection method and system for small oxygen sensor based on dynamic excitation

Abstract

The application discloses a small oxygen sensor portable detection method and system based on dynamic excitation, relates to the technical field of detection and measurement, and comprises the following steps: generating a synchronous data stream, constructing a synchronous factor sequence, calculating a posterior probability value, outputting an extraction starting point, mapping a time sequence into a high-dimensional phase space trajectory through phase space reconstruction, and calculating a maximum Lyapunov index. When a single parameter is qualified but the maximum Lyapunov index exceeds a preset early warning threshold, it is determined that there is potential failure. The application constructs a time sequence of a response time stage end voltage slope and an initial pump current fluctuation, maps the time sequence into a high-dimensional phase space trajectory through phase space reconstruction, calculates a maximum Lyapunov index, and performs monitoring. When a single parameter is qualified but the system is on the edge of instability, potential failure risks can be found in time. In combination with a change point extracted by an improved Bayesian online change point detection model as the starting point, the demand for high-precision, deep-level state perception and fault early warning of a small oxygen sensor is met.

Description

Technical Field

[0001] This invention relates to the field of detection and measurement technology, and in particular to a portable detection method and system for a small oxygen sensor based on dynamic excitation. Background Technology

[0002] In recent years, small oxygen sensors have become a core component in automotive emission control and industrial monitoring. Their performance directly affects the system's operating efficiency and safety. With the expansion of application scenarios, the demand for portable testing is increasing, requiring rapid and accurate assessment of sensor status in non-laboratory environments. Currently, most mainstream testing methods are based on the dynamic excitation principle. By collecting the sensor's output voltage and pump current signals, and combining them with variable-point detection algorithms such as Bayesian analysis, the aim is to improve detection accuracy through multi-dimensional signal characteristics to meet the high standards of modern industry for equipment health management.

[0003] Currently, Chinese invention patent CN116773738A discloses a portable nitrogen and oxygen sensor detector and detection method, including: an LCD display module, a main controller, a standard nitrogen and oxygen sensor, a CAN communication circuit, and a self-testing chamber structure for self-testing and calibrating the standard nitrogen and oxygen sensor mounted on a housing. In an air environment detection mode, the nitrogen oxide and oxygen concentrations are analyzed from the data collected by the nitrogen and oxygen sensor under test and compared with the set nitrogen oxide and oxygen concentration ranges to determine whether the nitrogen and oxygen sensor under test is functioning correctly. In a specific nitrogen and oxygen concentration atmosphere environment detection mode, the nitrogen oxide concentration analyzed from the data collected by the nitrogen and oxygen sensor under test is compared with the nitrogen oxide concentration analyzed from the data collected by the standard nitrogen and oxygen sensor to determine whether the nitrogen and oxygen sensor under test is functioning correctly. The results of the two test modes are used to determine whether the sensor's performance is abnormal, and human-computer interaction is also included for detecting the nitrogen and oxygen sensor under test. The handshake protocol for oxygen sensors is important, but current monitoring methods primarily focus on the pass / fail status of a single indicator, neglecting the nonlinear coupling relationships and complex dynamic behaviors between various parameters within the sensor's internal system. This means that while the sensor may pass individual parameter tests, the internal dynamic system may already be on the verge of instability, posing a potential failure risk. Such hidden dangers cannot be effectively identified in conventional testing. Existing change point detection and signal analysis techniques are usually based on linear assumptions or simple statistical characteristics, lacking the ability to mine the deep chaotic characteristics of time series and accurately characterize the system's evolution trajectory in high-dimensional phase space. Especially when facing noise interference and non-stationary signals common in portable testing, existing technologies struggle to accurately extract characteristic indicators reflecting the system's inherent stability from complex voltage and current data streams, leading to delayed or misjudgments of potential failures. This fails to meet the current demand for high-precision, deep-level state perception and fault early warning for small oxygen sensors. Summary of the Invention

[0004] The technical problem solved by this invention is that monitoring methods mainly focus on the pass / fail status of a single indicator, ignoring the nonlinear coupling relationships and complex dynamic behaviors between various parameters of the sensor's internal system. This leads to a situation where, although the sensor may pass the single-parameter test, the internal dynamic system may already be on the verge of instability, posing a potential risk of failure. Such hidden dangers cannot be effectively identified in conventional testing. Existing change point detection and signal analysis techniques are usually based on linear assumptions or simple statistical characteristics, lacking the ability to mine the deep chaotic characteristics of time series and failing to accurately characterize the evolution trajectory of the system in high-dimensional phase space. Especially when facing noise interference and non-stationary signals common in portable testing, existing technologies struggle to accurately extract characteristic indicators reflecting the essential stability of the system from complex voltage and current data streams, resulting in delayed or misjudgments of potential failures. This fails to meet the current demand for high-precision, deep-level state perception and fault early warning for small oxygen sensors.

[0005] To address the aforementioned technical problems, this invention provides the following technical solution: a portable detection method for a small oxygen sensor based on dynamic excitation, comprising the following steps: Step S1: Acquire the output voltage signal and pump current signal of the oxygen sensor and generate a synchronous data stream; Step S2: Construct a synchronization factor sequence based on the synchronization data stream, input the synchronization factor sequence into the improved Bayesian online change point detection model to calculate the posterior probability value, and output the change point when the posterior probability value is greater than the preset confidence threshold, and use the change point as the extraction starting point; Step S3: Extract the voltage slope at the end of the response time phase, and extract the pump current data segment starting from the extraction start point. Calculate the initial pump current fluctuation based on the pump current data segment. Construct a time series based on the voltage slope at the end of the response time phase and the initial pump current fluctuation. Map the time series to a high-dimensional phase space trajectory through phase space reconstruction. Calculate the maximum Lyapunov exponent of the high-dimensional phase space trajectory to obtain the current parameters of the system under test. If a single parameter is qualified but the maximum Lyapunov exponent exceeds a preset warning threshold, a potential failure is determined.

[0006] In a preferred embodiment of the portable detection method for a small oxygen sensor with dynamic excitation as described in this invention, step S1 specifically includes: The output voltage signal and pump current signal of the oxygen sensor are synchronously acquired at a preset fixed sampling frequency. The output voltage signal and the pump current signal are respectively subjected to moving average filtering to obtain the filtered voltage signal and the filtered current signal; The filtered voltage signal and the filtered current signal are respectively subjected to mean-reduction normalization to obtain the processed voltage signal and the processed current signal. Data points with the same timestamp in the processed voltage signal and processed current signal are time-aligned and arranged in chronological order according to their corresponding timestamps to generate a synchronous data stream.

[0007] In a preferred embodiment of the portable detection method for a small oxygen sensor with dynamic excitation as described in this invention, step S2 specifically includes: Extract the data points corresponding to the processed voltage signals from the synchronous data stream and arrange them in chronological order to generate a voltage sequence; Extract the data points corresponding to the processed current signals from the synchronous data stream and arrange them in chronological order to generate a current sequence; The voltage sequence and the current sequence are respectively subjected to Hilbert transform calculations to obtain voltage analytical signals and current analytical signals; The voltage analytical signal is subjected to arctangent calculation to obtain the instantaneous voltage phase sequence; The arctangent of the current analytical signal is calculated to obtain the instantaneous phase sequence of the current; Pair the elements with the same timestamp in the instantaneous voltage phase sequence with the instantaneous current phase sequence to generate phase data pairs; Calculate the difference between the instantaneous phase element of voltage and the instantaneous phase element of current in each phase data pair to obtain the instantaneous phase difference element; Arrange the instantaneous phase difference elements in chronological order to obtain the instantaneous phase difference sequence; The instantaneous phase difference sequence is weighted and averaged using an exponentially decaying sliding window function to generate a synchronization factor sequence. The synchronization factor sequence is input into the improved Bayesian online change point detection model to calculate the posterior probability value. When the posterior probability value is greater than the preset confidence threshold, the change point is output and used as the extraction starting point.

[0008] As a preferred embodiment of the portable detection method for a small oxygen sensor with dynamic excitation described in this invention, the step of inputting the synchronization factor sequence into the improved Bayesian online change point detection model to calculate the posterior probability value, and outputting the change point when the posterior probability value is greater than a preset confidence threshold, specifically includes: The synchronization factor sequence is input into the improved Bayesian online change point detection model. Based on the timestamp index corresponding to the synchronization factor sequence, the number of index interval steps from the current time to each historical time is calculated to generate the running time length. The running time length is substituted into the risk rate function of the improved Bayesian online change point detection model to calculate the change point transition probability value. Calculate the difference between constant 1 and the change point transition probability value to obtain the non-change point maintenance probability value; Obtain the historical prior probability value corresponding to the historical running time length of the previous moment; Multiply the invariant point maintenance probability value with the historical prior probability value to obtain the first prior transition value; Calculate the sum of all historical prior probability values ​​at the previous moment, and multiply the change point transition probability value by the sum to obtain the second prior transition value; Add the first prior transition value to the second prior transition value to obtain the current prior probability value corresponding to each running time length at the current time. Arrange the various running time lengths according to their numerical values, construct the mapping relationship between the running time lengths and the current prior probability values, and generate a prior probability distribution; Extract the data points corresponding to the processed voltage signal and the processed current signal from the synchronous data stream at the current moment; The data points corresponding to the processed voltage signal and the data points corresponding to the processed current signal are concatenated to obtain the observation vector at the current moment. The current observation vector is input into the likelihood function of the improved Bayesian online change point detection model to calculate the conditional probability of the current observation vector under each of the running time lengths. The prior probability value is multiplied by the corresponding conditional probability and then normalized to obtain the posterior probability value corresponding to each running time length. The posterior probability value is compared with a preset confidence threshold. When the posterior probability value is greater than the preset confidence threshold, the difference between the time index corresponding to the current moment and the running time length corresponding to the posterior probability value is calculated to obtain the change point time index, and the time point corresponding to the change point time index is determined as the change point. When all the posterior probability values ​​are less than or equal to the preset confidence threshold, it is determined that no change point has occurred at the current time.

[0009] In a preferred embodiment of the portable detection method for a small oxygen sensor with dynamic excitation as described in this invention, step S3 specifically includes: Extract the voltage slope at the end of the response time phase and calculate the initial pump current fluctuation after the extraction start point; A time series is constructed based on the voltage slope at the end of the response time phase and the initial pump current fluctuation, and the time series is mapped to a high-dimensional phase space trajectory through phase space reconstruction. Calculate the maximum Lyapunov exponent of the high-dimensional phase space trajectory; Obtain the current parameters of the system under test. When a single parameter is qualified but the maximum Lyapunov exponent exceeds the preset warning threshold, it is determined that there is a potential failure. The current parameters include the current output voltage, current response time, and current steady-state pump current; The determination of potential failure when a single parameter is qualified but the maximum Lyapunov exponent exceeds a preset warning threshold specifically includes: The current output voltage is compared with a preset acceptable range of output voltage to determine whether the current output voltage falls within the preset acceptable range of output voltage, and a first judgment result is generated. The current response time is compared with a preset acceptable response time range to determine whether the current response time falls within the preset acceptable response time range, and a second judgment result is generated. The current steady-state pump current is compared with the preset steady-state pump current qualified range to determine whether the current steady-state pump current falls within the preset steady-state pump current qualified range, and a third judgment result is generated. The maximum Lyapunov index is compared with a preset warning threshold to determine whether the maximum Lyapunov index is greater than or equal to the preset warning threshold, and a fourth judgment result is generated. When the first judgment result, the second judgment result, and the third judgment result are all qualified, and the fourth judgment result is yes, the judgment result that the system under test has a potential failure is output.

[0010] As a preferred embodiment of the portable detection method for a small oxygen sensor with dynamic excitation as described in this invention, the extraction of the voltage slope at the end of the response time phase specifically includes: In the synchronous data stream, a voltage data segment with a length equal to a preset first truncation length is extracted forward from the preset end time. The slope of the fitted line is obtained by performing a least-squares linear fitting calculation on the voltage data segment. Calculate the absolute value of the slope of the fitted straight line, and use the absolute value as the voltage slope at the end of the response time phase.

[0011] As a preferred embodiment of the portable detection method for a small oxygen sensor with dynamic excitation as described in this invention, the initial pump current fluctuation after calculating the extraction starting point specifically includes: In the synchronous data stream, starting from the extraction start point, a pump current data segment with a duration equal to a preset second extraction duration is extracted. Calculate the local variance of each data point within the pump current data segment, and calculate the arithmetic mean of each local variance to obtain the average local variance. The initial current fluctuation amplitude is obtained by taking the square root of the average local variance. The initial current disturbance frequency is obtained by counting the number of data points in the pump current data segment whose absolute value is greater than the preset current steady-state threshold. A feature vector is constructed by sequentially concatenating the initial current fluctuation amplitude and the initial current disturbance frequency, and the feature vector is used as the initial pump current fluctuation feature.

[0012] As a preferred embodiment of the portable detection method for a small oxygen sensor with dynamic excitation as described in this invention, the step of constructing a time series based on the end voltage slope of the response time stage and the initial pump current fluctuation, and mapping the time series into a high-dimensional phase space trajectory through phase space reconstruction, specifically includes: Starting from the extraction start point, voltage and current data segments of a preset analysis window length are extracted from the synchronous data stream. The instantaneous slope value of each data point in the voltage data segment and the local variance value of each data point in the current data segment at the corresponding time are concatenated according to the corresponding timestamp to generate a time series; The minimum embedding dimension is obtained by calculating the time series using the spurious nearest neighbor method. The optimal delay time is obtained by performing calculations on the time series using the mutual information method. Based on the minimum embedding dimension and the optimal delay time, the time series is reconstructed using delayed coordinate phase space to generate a high-dimensional phase space trajectory.

[0013] As a preferred embodiment of the portable detection method for a small oxygen sensor with dynamic excitation as described in this invention, the calculation of the maximum Lyapunov exponent of the high-dimensional phase space trajectory specifically includes: The initial phase point in the high-dimensional phase space trajectory is used as the reference phase point; The phase point with the smallest Euclidean distance to the reference phase point in the high-dimensional phase space trajectory is taken as the neighboring phase point; The reference phase point and the neighboring phase point are tracked and calculated along the time axis, and the Euclidean distance between the reference phase point and the neighboring phase point after each time step is calculated as the trajectory separation distance; When the trajectory separation distance is greater than the separation threshold during the tracking calculation process, the current tracking calculation process is taken as the evolution time period, and a new phase point is selected in the high-dimensional phase space trajectory to replace the neighboring phase point. The tracking calculation is then performed based on the replaced neighboring phase point to calculate the next evolution time period until the high-dimensional phase space trajectory has been traversed. For each evolution time period, the initial Euclidean distance between the reference phase point and the neighboring phase point at the start of evolution is obtained, and the final Euclidean distance between the reference phase point and the neighboring phase point at the end of evolution is obtained. Calculate the natural logarithm of the ratio of the terminating Euclidean distance to the initial Euclidean distance; Dividing the natural logarithm by the time step of the corresponding evolution period yields the Euclidean distance divergence rate of the evolution period. Calculate the average of the divergence rates of each Euclidean distance, and use it as the average divergence rate; The average divergence rate is taken as the maximum Lyapunov exponent.

[0014] A portable detection system for a small oxygen sensor based on dynamic excitation is applied in a portable detection method for a small oxygen sensor based on dynamic excitation. It includes an acquisition module, a calculation module, and a judgment module. The acquisition module is used to acquire the output voltage signal and pump current signal of the oxygen sensor and generate a synchronous data stream; The calculation module is used to construct a synchronization factor sequence based on the synchronization data stream, input the synchronization factor sequence into the improved Bayesian online change point detection model to calculate the posterior probability value, and output the change point when the posterior probability value is greater than the preset confidence threshold, and use the change point as the extraction starting point. The determination module is used to extract the voltage slope at the end of the response time phase, and extract the pump current data segment with the extraction start time as the starting time. Based on the pump current data segment, the initial pump current fluctuation is calculated. A time series is constructed based on the voltage slope at the end of the response time phase and the initial pump current fluctuation. The time series is mapped to a high-dimensional phase space trajectory through phase space reconstruction. The maximum Lyapunov exponent of the high-dimensional phase space trajectory is calculated to obtain the current parameters of the system under test. When a single parameter is qualified but the maximum Lyapunov exponent exceeds the preset warning threshold, a potential failure is determined.

[0015] The beneficial effects of this invention are as follows: By constructing a time series of the voltage slope at the end of the response time stage and the initial pump current fluctuation, and using phase space reconstruction to map it into a high-dimensional phase space trajectory, the maximum Lyapunov exponent is calculated. This effectively uncovers nonlinear coupling relationships and deep chaotic characteristics neglected by existing technologies, accurately characterizes the internal dynamic behavior of the system, and by monitoring the maximum Lyapunov exponent, potential failure risks can be detected in a timely manner when a single parameter is qualified but the system is on the edge of instability. This overcomes the limitations of traditional single-index monitoring. In addition, by combining the change point extracted by the improved Bayesian online change point detection model as the starting point, it can effectively cope with noise interference and non-stationary signals in portable detection, accurately extract characteristic indicators that reflect the essential stability of the system, thereby avoiding judgment lag or misjudgment, and meeting the needs for high-precision, deep-level state perception and fault early warning of small oxygen sensors. Attached Figure Description

[0016] Figure 1A schematic diagram of the basic process of a portable detection method for a small oxygen sensor based on dynamic excitation provided in an embodiment of the present invention; Figure 2 This is a basic flowchart of a portable detection system for a small oxygen sensor based on dynamic excitation, provided as an embodiment of the present invention. Detailed Implementation

[0017] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0018] Reference Figure 1 As an embodiment of the present invention, a portable detection method for a small oxygen sensor based on dynamic excitation is provided, comprising the following steps: Step S1: Acquire the output voltage signal and pump current signal of the oxygen sensor and generate a synchronous data stream; Step S2: Construct a synchronization factor sequence based on the synchronization data stream, input the synchronization factor sequence into the improved Bayesian online change point detection model to calculate the posterior probability value, and output the change point when the posterior probability value is greater than the preset confidence threshold, and use the change point as the extraction starting point; Step S3: Extract the voltage slope at the end of the response time phase, and extract the pump current data segment starting from the extraction start point. Calculate the initial pump current fluctuation based on the pump current data segment. Construct a time series based on the voltage slope at the end of the response time phase and the initial pump current fluctuation. Map the time series to a high-dimensional phase space trajectory through phase space reconstruction. Calculate the maximum Lyapunov exponent of the high-dimensional phase space trajectory to obtain the current parameters of the system under test. When a single parameter is qualified but the maximum Lyapunov exponent exceeds the preset warning threshold, a potential failure is determined.

[0019] This invention achieves spatiotemporal alignment of dual-source data by fusing output voltage and pump current signals to generate a synchronous data stream. Compared to existing technologies that rely solely on a single voltage signal, this provides richer data dimensions and significantly improved anti-interference capabilities. By utilizing a synchronization factor sequence combined with an improved Bayesian online change point detection model, it can accurately and adaptively capture abrupt changes in the sensor's response state, avoiding the misalignment of the extraction starting point caused by operating condition fluctuations in existing fixed threshold extraction methods. After change point extraction, it innovatively combines the end voltage slope of the response time stage with the initial pump current fluctuation to construct a time series, which is then mapped to a high-dimensional phase space trajectory to calculate the maximum Lyapunov exponent. This overcomes the limitation of existing technologies that rely solely on a single macroscopic parameter to determine pass / fail status. When conventional parameters are within acceptable limits but the maximum Lyapunov exponent exceeds the limit, a potential failure is identified, enabling this invention to provide early warning of microscopic dynamic instability and greatly reducing the risk of missed detections.

[0020] Step S1 specifically includes: The output voltage signal and pump current signal of the oxygen sensor are synchronously acquired at a preset fixed sampling frequency. The preset fixed sampling frequency is set to 1000 Hz, which is based on the frequency domain analysis of the sensor's dynamic response characteristics: First, the historical current change samples of the sensor under step excitation are extracted, and its spectrum distribution is obtained by fast Fourier transform. It is found that the highest frequency component of the effective signal is distributed within 400 Hz. In order to satisfy the Nyquist sampling theorem and retain sufficient time-domain waveform details, 1000 Hz is selected as the sampling frequency.

[0021] The output voltage signal and the pump current signal are respectively subjected to moving average filtering to obtain the filtered voltage signal and the filtered current signal; The filtered voltage signal and the filtered current signal are respectively subjected to mean-reduction and normalization processing to obtain the processed voltage signal and the processed current signal. Data points with the same timestamp in the processed voltage and current signals are time-aligned and arranged in chronological order according to their corresponding timestamps to generate a synchronous data stream.

[0022] In this embodiment, the moving average filtering uses an odd-numbered sliding window with a window length of 5. This window moves gradually with the time axis, calculating the arithmetic mean of all output voltage signals and pump current signals within the window. To avoid boundary signal distortion, a strategy of copying edge data to fill in the gaps at the beginning and end of the sequence is adopted. This arithmetic mean is used as the filtering result of the window center point, thereby using the low-pass filtering characteristics to filter out high-frequency random noise. The mean normalization process adopts the Z-score normalization algorithm. First, the overall average value of the filtered voltage signal and the filtered current signal is calculated. Then, the average value is subtracted from each data point in the signal, and then divided by the standard deviation of the signal sequence to obtain the processed voltage signal and processed current signal with a mean of zero and a variance of one. This eliminates the scale inconsistency caused by the difference in the range of different sensors. Same timestamp alignment refers to combining the processed voltage and processed current data points collected at the same time into binary data pairs based on the absolute time tag assigned by the high-precision timer during hardware acquisition. If a channel is missing current timestamp data due to sampling jitter, a linear interpolation algorithm is used to calculate the compensation value based on the data of adjacent times. After filling in the gaps, they are connected in chronological order to form a continuous synchronous data stream.

[0023] This invention sequentially performs moving average filtering and mean-reduction normalization on the original signal, effectively eliminating high-frequency noise and baseline drift interference. Compared to the coarse signal preprocessing methods in existing technologies, this scheme generates a synchronous data stream by aligning and sequentially arranging data points with the same timestamp, ensuring a strict correspondence between voltage and current on a microscopic time scale. This provides a high-fidelity data foundation for subsequent phase calculation and change-point detection, avoiding computational artifacts caused by asynchronous data.

[0024] Step S2 specifically includes: Extract the data points corresponding to the processed voltage signals from the synchronous data stream and arrange them in chronological order to generate a voltage sequence; Extract the data points corresponding to the current signals in the synchronous data stream and arrange them in chronological order to generate a current sequence; The voltage and current sequences are respectively subjected to Hilbert transform calculations to obtain the voltage analytical signal and the current analytical signal; The instantaneous phase sequence of the voltage is obtained by performing arctangent calculation on the voltage analytical signal; The arctangent of the current analytical signal is calculated to obtain the instantaneous phase sequence of the current; Pair elements with the same timestamp in the instantaneous voltage phase sequence with those in the instantaneous current phase sequence to generate phase data pairs; Calculate the difference between the instantaneous phase element of voltage and the instantaneous phase element of current in each phase data pair to obtain the instantaneous phase difference element; Arrange the instantaneous phase difference elements in chronological order to obtain the instantaneous phase difference sequence; A synchronization factor sequence is generated by using a weighted average of the instantaneous phase difference sequence with an exponentially decaying sliding window function. The synchronization factor sequence is input into the improved Bayesian online change point detection model to calculate the posterior probability value. When the posterior probability value is greater than the preset confidence threshold, the change point is output and used as the starting point for extracting steady-state stage features.

[0025] The preset reliability threshold is set at 0.95, which is based on the statistical analysis of historical change point detection data: First, the historical operating data is grouped into steady-state segments and change point segments, and the distribution of posterior probability values ​​at each time point is calculated. Then, the probability values ​​are sorted, and the minimum probability value that can cover 90% of the historical real change point samples is selected as the critical criterion. This value is calculated to be 0.95, thus ensuring the reliability of change point extraction.

[0026] In this embodiment, discrete Hilbert transforms are performed on the voltage and current sequences respectively. The original real signals are converted to the frequency domain using the Fast Fourier Transform algorithm, keeping the zero-frequency component unchanged. The amplitude of the positive frequency component is multiplied by 2, and the negative frequency component is set to zero. Then, the inverse Fourier transform is performed back to the time domain to obtain complex signals: the real part is the original signal, and the imaginary part is the orthogonal signal after the Hilbert transform. This complex signal is the analytic signal. The original real signal is used as the real part, and the orthogonal signal after the Hilbert transform is used as the imaginary part. These are combined to form the voltage analytic signal and the current analytic signal. Using the instantaneous phase calculation algorithm of discrete sequences, the imaginary part of the analytic signal is divided by the real part to obtain the arctangent of the four quadrants, resulting in the voltage instantaneous phase sequence and the current instantaneous phase sequence wrapped in the range from negative pi to positive pi. Then, the phase unwrapping algorithm is used to eliminate jumps. The unwrapped voltage phase is subtracted from the current phase at the same time stamp to obtain the instantaneous phase difference sequence. The exponentially decaying sliding window function employs a weighted moving average algorithm with a forgetting factor. The forgetting factor is set to 0.9, and data closer to the current time are given greater weight. The weight value decreases exponentially with the time interval. The weighted phase difference within the window is summed and divided by the total weight to obtain the average value, thereby generating the synchronization factor sequence for the current time, thus enhancing the ability of recent features to represent the current state.

[0027] This invention extracts the instantaneous phase of voltage and current using Hilbert transform and calculates the instantaneous phase difference sequence. Compared to existing techniques that directly employ time-domain amplitude features, the phase difference feature can reveal the dynamic synchronicity of electrochemical reactions within the sensor at a deeper level. A synchronization factor sequence is generated by weighted averaging using an exponentially decaying sliding window function, giving higher weight to recent phase changes. This makes the synchronization factor sequence extremely sensitive to subtle early degradation of the sensor state, significantly improving the sensitivity for extracting potential fault features.

[0028] The synchronization factor sequence is input into the improved Bayesian online change point detection model to calculate the posterior probability value. When the posterior probability value is greater than the preset confidence threshold, the specific change points are output as follows: The synchronization factor sequence is input into the improved Bayesian online change point detection model. Based on the timestamp index corresponding to the synchronization factor sequence, the number of index interval steps from the current time to each historical time is calculated, and the running time length is generated. The runtime length is substituted into the risk rate function of the improved Bayesian online change point detection model to calculate the change point transition probability value. Calculate the difference between constant 1 and the transition probability value of the change point to obtain the maintenance probability value of the non-change point; Obtain the historical prior probability value corresponding to the historical running time length of the previous moment; Multiply the invariant point maintenance probability value with the historical prior probability value to obtain the first prior transition value; Calculate the sum of all historical prior probability values ​​from the previous moment, and multiply the change point transition probability value by the sum to obtain the second prior transition value; Add the first prior transition value to the second prior transition value to obtain the current prior probability value corresponding to each running time length at the current moment; Arrange the various runtime lengths according to their numerical values, construct a mapping relationship between runtime lengths and current prior probability values, and generate a prior probability distribution; Extract the data points corresponding to the processed voltage signal and the processed current signal from the synchronous data stream at the current moment; The data points corresponding to the processed voltage signal and the data points corresponding to the processed current signal are concatenated to obtain the observation vector at the current moment. Input the current observation vector into the likelihood function of the improved Bayesian online change point detection model, and calculate the conditional probability of the current observation vector at each running time length; The prior probability value is multiplied by the corresponding conditional probability and then normalized to obtain the posterior probability value corresponding to each running time length. Compare the posterior probability value with a pre-set reliability threshold; When the posterior probability value is greater than the preset confidence threshold, calculate the difference between the time index corresponding to the current time and the running time length corresponding to the posterior probability value to obtain the change point time index, and determine the time point corresponding to the change point time index as the change point; When all posterior probability values ​​are less than or equal to the preset confidence threshold, it is determined that no change point has occurred at the current time.

[0029] The improved Bayesian online change point detection model is based on the Bayesian inference framework and the Markov assumption. It realizes the real-time and accurate location of the change point by recursively calculating the posterior probability of the state change point at the current moment. The core of the model is to combine the historical prior probability distribution with the likelihood function of the current observation data. The running time is defined as the number of sampling points traversed from the current moment to the previous possible change point, which represents the duration for which the system maintains the current physical state.

[0030] The risk rate function uses an exponential distribution function algorithm with parameter lambda to calculate the probability value of a state change at the next moment, i.e., the transition probability value, given that the system has been running for a specific duration. The non-change point maintenance probability value is calculated as a constant 1 minus this transition probability value. The calculation of the prior probability distribution combines the continuation term of the previous moment's historical prior probability multiplied by the non-change point maintenance probability value, and the mutation term of the sum of historical prior probabilities multiplied by the change point transition probability value, forming the prediction step of the Markov chain, thereby accurately characterizing the evolutionary transition mechanism of the system's state maintenance or mutation.

[0031] The likelihood function adopts a Gaussian distribution and uses the expectation-maximization algorithm to update the mean vector and covariance matrix of the current observation vector online. By calculating the Gaussian probability density of the current voltage and current concatenated observation vector over a given historical running length, the degree of matching between the current observation data and the historical state distribution is characterized. In the update step, the prior probability obtained in the prediction step is multiplied by the conditional probability calculated by the likelihood function. Then, the product over all running lengths is summed and normalized to obtain the posterior probability value corresponding to each running length. When the posterior probability exceeds a preset confidence threshold, a change point is determined, and the difference between the running length corresponding to the posterior probability and the current time index is calculated to obtain the change point time index, which is used as the extraction starting point. This model, through a continuously iterative prediction and update mechanism, achieves adaptive online capture of abrupt changes in the sensor's dynamic response state, without requiring offline training of global data. This greatly suppresses false alarms of change points caused by transient noise and ensures the absolute reliability of the extraction starting point.

[0032] In this embodiment, the running time refers to the number of sampling points from the current moment to the previous possible change point, representing the duration for which the system maintains its current state. The risk rate function uses an exponential distribution function algorithm with a parameter of lambda, representing the probability value of a change in state at the current moment, i.e., the change point transition probability value. The non-change point maintenance probability value is a constant 1 minus this transition probability value. The calculation of the prior probability distribution combines the historical prior probability of the previous moment multiplied by the continuation term of the non-change point maintenance probability value, and the sum of historical prior probabilities multiplied by the change point transition probability value, forming the prediction step of the Markov chain. The likelihood function adopts a Gaussian distribution form, and the mean and covariance matrix of the current observation vector are updated online using the expectation-maximization algorithm. The conditional probability of the current observation vector under a given historical running time is calculated. After multiplying the prior probability and the conditional probability, the sum of the products under all running time lengths is used as the denominator for normalization processing to obtain the posterior probability value corresponding to each running time length. When the posterior probability exceeds the preset confidence threshold, it is determined that a change point has occurred.

[0033] This invention deeply reconstructs the specific computational process of the improved Bayesian online change point detection model. By introducing the runtime length combined with a risk rate function to calculate the change point transition probability and the non-change point maintenance probability, it accurately characterizes the Markov characteristics of sensor state evolution. Compared with the static window change point detection method in the prior art, this scheme achieves online real-time updating of the posterior probability through joint likelihood calculation of historical prior probabilities and the current observation vector. This not only significantly reduces the dependence on prior knowledge but also greatly suppresses false change point alarms caused by transient noise, ensuring the absolute reliability of the extracted starting point.

[0034] Step S3 specifically includes: Extract the voltage slope at the end of the response time phase and calculate the initial pump current fluctuation after the extraction start point; A time series is constructed based on the voltage slope at the end of the response time stage and the initial pump current fluctuation, and the time series is mapped to a high-dimensional phase space trajectory through phase space reconstruction. Calculate the maximum Lyapunov exponent of a trajectory in a high-dimensional phase space; Obtain the current parameters of the system under test. When a single parameter is qualified but the maximum Lyapunov exponent exceeds the preset warning threshold, it is determined that there is a potential failure. The preset warning threshold is set to 0.05, which is based on the dynamic statistical analysis of historical potential failure samples: First, the phase space trajectory data of the sensor in the critical state between normal aging and potential failure is extracted, and its maximum Lyapunov exponent sequence is calculated. The exponent values ​​are sorted by size, and the smallest exponent value that can cover 85% of historical early failure samples is selected as the warning limit. The value is 0.05.

[0035] Current parameters include current output voltage, current response time, and current steady-state pump current; When a single parameter is qualified but the maximum Lyapunov exponent exceeds the preset warning threshold, a potential failure is determined, specifically including: The current output voltage is compared with the preset acceptable range of output voltage to determine whether the current output voltage falls within the preset acceptable range of output voltage, and a first judgment result is generated. The preset acceptable output voltage range is set to 0.45 volts to 0.55 volts. This range is established based on the statistical analysis of historical acceptable steady-state output data from sensors: First, 500 samples of steady-state output voltage under normal operating conditions are collected and their mean and standard deviation are calculated. The mean is used as the center, and plus or minus three times the standard deviation is taken as the confidence interval covering 99.7% of the normal samples. The boundary of this interval is determined to be 0.45 volts to 0.55 volts based on the statistical analysis of the data.

[0036] The current response time is compared with the preset acceptable response time range to determine whether the current response time falls within the preset acceptable response time range, and a second judgment result is generated. The preset acceptable response time range is set to 50 milliseconds to 120 milliseconds. This range is based on the statistical analysis of historical dynamic excitation test data. First, the time samples of historical qualified sensors transitioning from the initial value to the steady-state value under step excitation are extracted. The response time values ​​of each group are sorted, and the upper and lower bounds of the time that can cover 90% of the normal response samples are selected as the qualified judgment limits. According to statistics, this range is 50 milliseconds to 120 milliseconds.

[0037] The current steady-state pump current is compared with the preset steady-state pump current qualified range to determine whether the current steady-state pump current falls within the preset steady-state pump current qualified range, and a third judgment result is generated. The preset acceptable range for steady-state pump current is set to 3 mA to 8 mA. This is based on the statistical analysis of pump current data under normal operating conditions in history. First, samples of steady-state pump current after the sensor reaches thermal equilibrium are extracted. The current density distribution of each sample is calculated. The samples are sorted by current value. The current range that covers the first 85% of the normal operating samples is selected as the criterion. This range is calculated to be 3 mA to 8 mA.

[0038] The maximum Lyapunov index is compared with the preset warning threshold to determine whether the maximum Lyapunov index is greater than or equal to the preset warning threshold, and a fourth judgment result is generated. When the first, second, and third judgment results are all qualified, and the fourth judgment result is yes, the system under test is output as a judgment result indicating that there is a potential failure.

[0039] When at least one of the first, second, and third judgment results is unqualified, or when the first, second, and third judgment results are all qualified and the fourth judgment result is negative, the judgment result that the system under test is operating normally is output.

[0040] The potential failure determination logic in this embodiment of the invention is mainly used to provide early warning of potential dynamic divergence within the system by leveraging the coupling feature of the maximum Lyapunov exponent when all external single parameters of the system under test are normal.

[0041] Specifically, the condition monitoring of the system under test is divided into two levels: explicit fault monitoring and potential failure early warning. This step is only responsible for potential failure warning and not for obvious failure judgment. The output has only two states: potential failure exists and potential failure does not exist. When the normal parameters are not qualified, the obvious failure has already alarmed. This step returns to normal operation and does not alarm.

[0042] Obvious fault level: When any of the output voltage, response time, or steady-state pump current of the system under test falls into the unqualified range, it indicates that the system has experienced a clear and substantial functional abnormality, including voltage drop and response hysteresis. At this time, the conventional control unit of the system will directly trigger an obvious fault alarm and take protective action without the intervention of the potential failure judgment module of this invention, that is, it will not output the potential failure judgment result.

[0043] Potential failure level: When all three single parameters mentioned above fall within the qualified range, that is, the surface operation is normal, but the calculated maximum Lyapunov exponent is greater than or equal to the warning threshold, it indicates that although the external performance of the system is normal, its internal state trajectory has shown an exponential divergence trend and is in an unstable critical state. At this time, the output shows a judgment result of potential failure to prompt preventive maintenance.

[0044] If all three individual parameters are qualified and the maximum Lyapunov exponent is less than the warning threshold, it indicates that the internal and external states of the system are stable, and the output judgment result indicates that the system is operating normally.

[0045] Through the above-mentioned hierarchical judgment logic, the present invention can effectively identify potential divergence risks within the system before traditional single-parameter alarms, thus achieving early warning of faults.

[0046] In this embodiment, the current parameters include the current output voltage, current response time, and current steady-state pump current. These three parameters are conventional macroscopic indicators used in the prior art to measure the performance of oxygen sensors. The first, second, and third judgment results all characterize whether the sensor meets the standards at the conventional functional level. The maximum Lyapunov exponent reflects the degree of chaos within the system. When this exponent is greater than or equal to a preset warning threshold, it indicates that the microscopic dynamic trajectory within the system has diverged. The judgment logic of this invention is a strict combination of preconditions. A potential failure judgment is only output if the maximum Lyapunov exponent still exceeds the limit, provided that all three conventional parameters fall within the acceptable range. This logic eliminates interference from dynamic changes in the system caused by unacceptable macroscopic parameters, and specifically targets early warning of latent degradation.

[0047] This invention constructs a multi-dimensional joint failure judgment logic, using the maximum Lyapunov exponent as the core criterion, and cross-validating it with the current output voltage, response time, and steady-state pump current. Existing technologies typically only issue alarms when conventional parameters exceed acceptable ranges, at which point the sensor has already suffered substantial damage. This invention, however, can keenly capture the divergent trends of the system's internal dynamic trajectory even when conventional parameters appear acceptable, identifying potential failures in advance. This represents a leap from reactive intervention to proactive prediction, significantly improving the reliability and early warning capabilities of portable detection.

[0048] The extraction of the voltage slope at the end of the response time phase specifically includes: In a synchronous data stream, a voltage data segment with a length equal to a preset first truncation length is extracted forward from the preset end time. The preset end time is set at 300 milliseconds after the step excitation is applied. The basis for this is the statistical analysis of historical sensor response recovery data: First, extract the time sample required for the voltage to reach 98% of the final steady state value from the historical step response data. Sort the time values ​​of each sample and select the upper bound of the time that can cover 85% of the historical fast response samples as the termination node of the response time stage. According to statistics, this time node is 300 milliseconds.

[0049] The preset first truncation length is set to 50 sampling points. The basis for this setting is the statistical analysis of the voltage slope characteristic data at the end of the historical response: First, the voltage sequence before the end of the historical normal response stage is extracted, and the variance stability of the fitted slope under different truncation lengths is calculated. The variance values ​​are sorted according to the truncation length, and the shortest length that can cover 80% of the high fit samples is selected as the critical criterion. This value corresponds to 50 sampling points.

[0050] The slope of the fitted line is obtained by performing least-squares linear fitting on the voltage data segment. Calculate the absolute value of the slope of the fitted straight line and use the absolute value as the voltage slope at the end of the response time phase.

[0051] In this embodiment, a voltage data segment with a preset end time as the endpoint is extracted forward by a length equal to the preset first extraction length. This extraction length is set according to the sensor response characteristics to be sufficient to reflect the voltage change trend at the end of the period. Least squares linear fitting is used with the timestamp as the independent variable and the voltage value of the data point as the dependent variable to construct a linear equation. By minimizing the sum of squares of the vertical distances from each data point to the line, the slope value of the line is solved. The absolute value of the slope value is taken to eliminate the influence of positive and negative signs, and it is used as the voltage slope at the end of the response time stage, which characterizes the rate at which the voltage tends to stabilize.

[0052] This invention extracts the voltage slope at the end of the response time phase by truncating voltage data segments of a specific length and performing least-squares linear fitting. Compared to the crude approach of directly taking the voltage difference at a fixed moment in existing technologies, this scheme uses the absolute value of the fitted slope to characterize the voltage change rate at the end of the response, effectively smoothing out random jitter in the terminal data and more scientifically reflecting the polarization state of the sensor before reaching steady state, thus providing high-quality feature components for subsequent phase space reconstruction.

[0053] The calculation of the initial pump current fluctuation after the starting point is specifically included in: In the synchronous data stream, the pump current data segment with the extraction start time as the starting time is extracted, and the extraction duration is equal to the preset second extraction duration. The second interception duration is preset to 100 milliseconds. The basis for this is the statistical analysis of the initial current disturbance data after the historical change point: First, extract the duration sample of the weak fluctuation of current after the historical change point occurs, sort the duration data by value, select the shortest duration that can cover the first 85% of the effective disturbance samples as the feature extraction window, and statistically set this duration to 100 milliseconds.

[0054] Calculate the local variance of each data point within the pump current data segment, and calculate the arithmetic mean of each local variance to obtain the average local variance. The initial current fluctuation amplitude is obtained by taking the square root of the average local variance. The frequency of initial current disturbance is obtained by counting the number of data points whose absolute value is greater than the preset steady-state threshold within the pump current data segment. The preset steady-state current threshold is set to 0.2 mA. The basis for this setting is the statistical analysis of historical steady-state current noise amplitude data. First, the baseline noise sample of the sensor current under absolute steady state is extracted and its fluctuation amplitude is calculated. The amplitude data is sorted, and the maximum fluctuation amplitude that can cover 95% of the steady-state noise sample is selected as the critical value to distinguish between normal fluctuations and abnormal disturbances. This value is calculated to be 0.2 mA.

[0055] A feature vector is constructed by sequentially concatenating the initial current fluctuation amplitude and the initial current disturbance frequency, and the feature vector is used as the feature of the initial pump current fluctuation.

[0056] In this embodiment, a sliding window mechanism is used to calculate the local variance within the pump current data segment extracted from the starting point. A fixed number of adjacent points are selected as the center of each data point to form a local window, and the variance of the data within this window is calculated as the local variance of the current data point. The local variances of all points within the data segment are summed and squared to obtain the initial current fluctuation amplitude, reflecting the energy magnitude of the current fluctuation. A preset current steady-state threshold is a critical value for determining whether the current is in a steady state. The number of data points in the data segment whose absolute value is greater than this threshold is counted as the initial current disturbance frequency, reflecting the density of the fluctuation. The initial current fluctuation amplitude is used as the first element, and the initial current disturbance frequency is used as the second element. They are arranged in order to form a two-dimensional feature vector, which is the initial pump current fluctuation characteristic.

[0057] This invention, when extracting initial pump current fluctuations, not only obtains the fluctuation amplitude by summing and taking the square root of local variances, but also obtains the disturbance frequency by counting the number of out-of-limit data points. Compared with existing technologies that only calculate the overall variance, this invention concatenates the amplitude and frequency into a feature vector, comprehensively characterizing the current perturbation characteristics after the extraction starting point from both energy and frequency dimensions. This completely avoids the problem of missing high-frequency, low-amplitude disturbances by a single amplitude index, significantly improving the accuracy of characterizing minute abnormal currents.

[0058] A time series is constructed based on the voltage slope at the end of the response time phase and the initial pump current fluctuation, and the time series is mapped to a high-dimensional phase space trajectory through phase space reconstruction, specifically including: Starting from the extraction start point, voltage and current data segments of a preset analysis window length are extracted from the synchronous data stream. The instantaneous slope value of each data point in the voltage data segment and the local variance value of each data point in the current data segment at the corresponding time are concatenated according to the corresponding timestamp to generate a multidimensional time series. The spurious nearest neighbor method is used to calculate the minimum embedding dimension of the time series. The optimal delay time is obtained by performing calculations on the time series using the mutual information method. Based on the minimum embedding dimension and the optimal delay time, the time series is reconstructed in delayed coordinate phase space to generate a high-dimensional phase space trajectory.

[0059] In this embodiment, the false nearest neighbor method calculates the change in distance between the nearest neighbors of each phase point in the phase space as the embedding dimension increases by gradually increasing the dimension. If the distance increases significantly, the nearest neighbor is determined to be a false nearest neighbor. As the dimension increases, the proportion of false nearest neighbors gradually decreases. When this proportion drops to near zero or less than a set minimum value, the corresponding dimension is the minimum embedding dimension. The mutual information method calculates the mutual information of the time series at different time delays and plots a curve of mutual information changing with the delay time. When the curve reaches its first minimum point, the delay time corresponding to that point is the optimal delay time. Based on these two parameters, starting from the first data point of the time series, a data point is taken every optimal delay time, and several data points with the minimum embedding dimension are continuously taken to form a multidimensional coordinate, which is considered as a phase point. Multiple phase points are generated by traversing the entire time series and connected to form a high-dimensional phase space trajectory.

[0060] In the phase space reconstruction process, this invention abandons the empirical value setting method commonly used in existing technologies and innovatively adopts the spurious nearest neighbor method to calculate the minimum embedding dimension, and uses the mutual information method to calculate the optimal delay time. This data-driven adaptive parameter selection mechanism ensures that the high-dimensional phase space trajectory can restore the topology of the prime mover system to the greatest extent and without distortion, avoiding trajectory distortion caused by trajectory overlap due to excessively low embedding dimension or inappropriate delay time, thus laying a solid foundation for the accurate calculation of the Lyapunov exponent.

[0061] Calculating the maximum Lyapunov exponent of a high-dimensional phase space trajectory specifically includes: The initial phase point in the trajectory of the high-dimensional phase space is used as the reference phase point; The first phase point in the high-dimensional phase space trajectory is taken as the initial reference phase point, and the reference phase point is gradually advanced along the time axis. The phase point with the smallest Euclidean distance from the reference phase point in the high-dimensional phase space trajectory is taken as the neighboring phase point; The reference phase point and the neighboring phase points are tracked and calculated along the time axis, and the Euclidean distance between the reference phase point and the neighboring phase points after each time step is calculated as the trajectory separation distance. When the trajectory separation distance exceeds the separation threshold during the tracking calculation process, the current tracking calculation process is taken as the evolution time period. A new phase point is selected in the high-dimensional phase space trajectory to replace the neighboring phase point. The tracking calculation is then performed based on the replaced neighboring phase point to calculate the next evolution time period until the high-dimensional phase space trajectory has been traversed. For each evolution time period, obtain the initial Euclidean distance between the reference phase point and the neighboring phase points at the start of the evolution, and obtain the final Euclidean distance between the reference phase point and the neighboring phase points at the end of the evolution. Calculate the natural logarithm of the ratio of the terminating Euclidean distance to the initial Euclidean distance; Dividing the natural logarithm by the time step of the corresponding evolution period yields the Euclidean distance divergence rate for the evolution period. Calculate the average of the divergence rates of each Euclidean distance, and use it as the average divergence rate; The average divergence rate is taken as the maximum Lyapunov exponent.

[0062] In this embodiment, the initial phase point is the first phase point in the high-dimensional phase space trajectory. The phase point with the smallest Euclidean distance, excluding the reference phase point itself, is searched as a neighboring phase point. The reference phase point and neighboring phase points are traced forward along the time axis, and the Euclidean distance between them at each time step is calculated as the trajectory separation distance. The separation threshold is a preset maximum allowable separation distance. When the trajectory separation distance exceeds this threshold, the current evolution time period ends. To avoid nonlinear saturation, a new phase point with the smallest distance is searched near the reference phase point to replace the original neighboring phase point, and the tracing of the next evolution time period begins, until all phase points have been traversed. Within each evolution time period, the initial Euclidean distance at the start time and the final Euclidean distance at the end time are recorded. The natural logarithm of the ratio of the final distance to the initial distance is calculated and divided by the total time step of that time period to obtain the divergence rate for that time period. The average divergence rate of all evolution time periods is calculated as the average divergence rate of a reference phase point. The maximum value among the average divergence rates corresponding to all reference phase points is selected as the maximum Lyapunov exponent.

[0063] This invention introduces a separation threshold and a dynamic replacement mechanism for neighboring phase points when calculating the maximum Lyapunov exponent. When the trajectory separation distance between the reference phase point and neighboring phase points exceeds the separation threshold, a new phase point is promptly selected to replace the neighboring phase point for the next evolution time period calculation. Compared to the nonlinear saturation problem caused by fixing neighboring phase points throughout the process in existing technologies, this scheme effectively avoids divergence distortion in phase space trajectory tracking. By calculating the Euclidean distance divergence rate piecewise and taking the maximum value, the maximum divergence rate of the system trajectory is accurately quantified, ensuring the high robustness of the potential failure warning index.

[0064] Reference Figure 2 This is another embodiment of the present invention. This embodiment differs from the first embodiment in that it provides a portable detection system for a small oxygen sensor based on dynamic excitation, including a data acquisition module, a calculation module and a judgment module. The acquisition module is used to acquire the output voltage signal and pump current signal of the oxygen sensor and generate a synchronous data stream; The calculation module is used to construct a synchronization factor sequence based on the synchronization data stream. The synchronization factor sequence is input into the improved Bayesian online change point detection model to calculate the posterior probability value. When the posterior probability value is greater than the preset confidence threshold, the change point is output and the change point is used as the extraction starting point. The judgment module is used to extract the voltage slope at the end of the response time stage and extract the pump current data segment with the extraction start time as the starting time. Based on the pump current data segment, the initial pump current fluctuation is calculated. A time series is constructed based on the voltage slope at the end of the response time stage and the initial pump current fluctuation. The time series is mapped to a high-dimensional phase space trajectory through phase space reconstruction. The maximum Lyapunov exponent of the high-dimensional phase space trajectory is calculated to obtain the current parameters of the system under test. When a single parameter is qualified but the maximum Lyapunov exponent exceeds the preset warning threshold, a potential failure is judged.

[0065] This invention constructs a time series of the voltage slope at the end of the response time phase and the initial pump current fluctuation, and maps it to a high-dimensional phase space trajectory using phase space reconstruction. It then calculates the maximum Lyapunov exponent, effectively uncovering nonlinear coupling relationships and deep chaotic characteristics neglected by existing technologies. This allows for precise characterization of the system's internal dynamics. By monitoring the maximum Lyapunov exponent, potential failure risks can be detected promptly when a single parameter is within acceptable limits but the system is on the verge of instability, overcoming the limitations of traditional single-indicator monitoring. Furthermore, by combining the improved Bayesian online change point detection model with the extracted change point as the starting point, it effectively addresses noise interference and non-stationary signals in portable detection, accurately extracting characteristic indicators reflecting the system's inherent stability. This avoids judgment lag or misjudgment, meeting the demand for high-precision, deep-level state perception and fault early warning for small oxygen sensors.

[0066] The various embodiments of the techniques described above can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0067] The program code used to implement the methods of the present invention can be written in any combination of one or more programming languages. This program code can be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code can be executed entirely on the machine, partially on the machine, as a standalone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0068] In the context of this invention, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

Claims

1. A portable detection method for a small oxygen sensor based on dynamic excitation, characterized in that, Includes the following steps: Step S1: Acquire the output voltage signal and pump current signal of the oxygen sensor and generate a synchronous data stream; Step S2: Construct a synchronization factor sequence based on the synchronization data stream, input the synchronization factor sequence into the improved Bayesian online change point detection model to calculate the posterior probability value, and output the change point when the posterior probability value is greater than the preset confidence threshold, and use the change point as the extraction starting point; Step S3: Extract the voltage slope at the end of the response time phase, and extract the pump current data segment starting from the extraction start point. Calculate the initial pump current fluctuation based on the pump current data segment. Construct a time series based on the voltage slope at the end of the response time phase and the initial pump current fluctuation. Map the time series to a high-dimensional phase space trajectory through phase space reconstruction. Calculate the maximum Lyapunov exponent of the high-dimensional phase space trajectory to obtain the current parameters of the system under test. If a single parameter is qualified but the maximum Lyapunov exponent exceeds a preset warning threshold, a potential failure is determined.

2. The portable detection method for a dynamically excited miniature oxygen sensor as described in claim 1, characterized in that, Step S1 specifically includes: The output voltage signal and pump current signal of the oxygen sensor are synchronously acquired at a preset fixed sampling frequency. The output voltage signal and the pump current signal are respectively subjected to moving average filtering to obtain the filtered voltage signal and the filtered current signal; The filtered voltage signal and the filtered current signal are respectively subjected to mean-reduction normalization to obtain the processed voltage signal and the processed current signal. Data points with the same timestamp in the processed voltage signal and processed current signal are time-aligned and arranged in chronological order according to their corresponding timestamps to generate a synchronous data stream.

3. The portable detection method for a dynamically excited miniature oxygen sensor as described in claim 2, characterized in that, Step S2 specifically includes: Extract the data points corresponding to the processed voltage signals from the synchronous data stream and arrange them in chronological order to generate a voltage sequence; Extract the data points corresponding to the processed current signals from the synchronous data stream and arrange them in chronological order to generate a current sequence; The voltage sequence and the current sequence are respectively subjected to Hilbert transform calculations to obtain voltage analytical signals and current analytical signals; The voltage analytical signal is subjected to arctangent calculation to obtain the instantaneous voltage phase sequence; The arctangent of the current analytical signal is calculated to obtain the instantaneous phase sequence of the current; Pair the elements with the same timestamp in the instantaneous voltage phase sequence with the instantaneous current phase sequence to generate phase data pairs; Calculate the difference between the instantaneous phase element of voltage and the instantaneous phase element of current in each phase data pair to obtain the instantaneous phase difference element; Arrange the instantaneous phase difference elements in chronological order to obtain the instantaneous phase difference sequence; The instantaneous phase difference sequence is weighted and averaged using an exponentially decaying sliding window function to generate a synchronization factor sequence. The synchronization factor sequence is input into the improved Bayesian online change point detection model to calculate the posterior probability value. When the posterior probability value is greater than the preset confidence threshold, the change point is output and used as the extraction starting point.

4. The portable detection method for a dynamically excited miniature oxygen sensor as described in claim 3, characterized in that, The step of inputting the synchronization factor sequence into the improved Bayesian online change point detection model to calculate the posterior probability value, and outputting the change point when the posterior probability value is greater than a preset confidence threshold, specifically includes: The synchronization factor sequence is input into the improved Bayesian online change point detection model. Based on the timestamp index corresponding to the synchronization factor sequence, the number of index interval steps from the current time to each historical time is calculated to generate the running time length. The running time length is substituted into the risk rate function of the improved Bayesian online change point detection model to calculate the change point transition probability value. Calculate the difference between constant 1 and the change point transition probability value to obtain the non-change point maintenance probability value; Obtain the historical prior probability value corresponding to the historical running time length of the previous moment; Multiply the invariant point maintenance probability value with the historical prior probability value to obtain the first prior transition value; Calculate the sum of all historical prior probability values ​​at the previous moment, and multiply the change point transition probability value by the sum to obtain the second prior transition value; Add the first prior transition value to the second prior transition value to obtain the current prior probability value corresponding to each running time length at the current time. Arrange the various running time lengths according to their numerical values, construct the mapping relationship between the running time lengths and the current prior probability values, and generate a prior probability distribution; Extract the data points corresponding to the processed voltage signal and the processed current signal from the synchronous data stream at the current moment; The data points corresponding to the processed voltage signal and the data points corresponding to the processed current signal are concatenated to obtain the observation vector at the current moment. The current observation vector is input into the likelihood function of the improved Bayesian online change point detection model to calculate the conditional probability of the current observation vector under each of the running time lengths. The prior probability value is multiplied by the corresponding conditional probability and then normalized to obtain the posterior probability value corresponding to each running time length. The posterior probability value is compared with a preset confidence threshold. When the posterior probability value is greater than the preset confidence threshold, the difference between the time index corresponding to the current moment and the running time length corresponding to the posterior probability value is calculated to obtain the change point time index, and the time point corresponding to the change point time index is determined as the change point. When all the posterior probability values ​​are less than or equal to the preset confidence threshold, it is determined that no change point has occurred at the current time.

5. The portable detection method for a dynamically excited miniature oxygen sensor as described in claim 4, characterized in that, Step S3 specifically includes: Extract the voltage slope at the end of the response time phase and calculate the initial pump current fluctuation after the extraction start point; A time series is constructed based on the voltage slope at the end of the response time phase and the initial pump current fluctuation, and the time series is mapped to a high-dimensional phase space trajectory through phase space reconstruction. Calculate the maximum Lyapunov exponent of the high-dimensional phase space trajectory; Obtain the current parameters of the system under test. When a single parameter is qualified but the maximum Lyapunov exponent exceeds the preset warning threshold, it is determined that there is a potential failure. The current parameters include the current output voltage, current response time, and current steady-state pump current; The determination of potential failure when a single parameter is qualified but the maximum Lyapunov exponent exceeds a preset warning threshold specifically includes: The current output voltage is compared with a preset acceptable range of output voltage to determine whether the current output voltage falls within the preset acceptable range of output voltage, and a first judgment result is generated. The current response time is compared with a preset acceptable response time range to determine whether the current response time falls within the preset acceptable response time range, and a second judgment result is generated. The current steady-state pump current is compared with the preset steady-state pump current qualified range to determine whether the current steady-state pump current falls within the preset steady-state pump current qualified range, and a third judgment result is generated. The maximum Lyapunov index is compared with a preset warning threshold to determine whether the maximum Lyapunov index is greater than or equal to the preset warning threshold, and a fourth judgment result is generated. When the first judgment result, the second judgment result, and the third judgment result are all qualified, and the fourth judgment result is yes, the judgment result that the system under test has a potential failure is output.

6. The portable detection method for a dynamically excited miniature oxygen sensor as described in claim 5, characterized in that, The voltage slope at the end of the extracted response time phase specifically includes: In the synchronous data stream, a voltage data segment with a length equal to a preset first truncation length is extracted forward from the preset end time. The slope of the fitted line is obtained by performing a least-squares linear fitting calculation on the voltage data segment. Calculate the absolute value of the slope of the fitted straight line, and use the absolute value as the voltage slope at the end of the response time phase.

7. The portable detection method for a dynamically excited miniature oxygen sensor as described in claim 6, characterized in that, The initial pump current fluctuation after the calculation of the starting point specifically includes: In the synchronous data stream, starting from the extraction start point, a pump current data segment with a duration equal to a preset second extraction duration is extracted. Calculate the local variance of each data point within the pump current data segment, and calculate the arithmetic mean of each local variance to obtain the average local variance. The initial current fluctuation amplitude is obtained by taking the square root of the average local variance. The initial current disturbance frequency is obtained by counting the number of data points in the pump current data segment whose absolute value is greater than the preset current steady-state threshold. A feature vector is constructed by sequentially concatenating the initial current fluctuation amplitude and the initial current disturbance frequency, and the feature vector is used as the initial pump current fluctuation feature.

8. The portable detection method for a dynamically excited miniature oxygen sensor as described in claim 7, characterized in that, The process of constructing a time series based on the end voltage slope and initial pump current fluctuation of the response time phase, and mapping the time series into a high-dimensional phase space trajectory through phase space reconstruction, specifically includes: Starting from the extraction start point, voltage and current data segments of a preset analysis window length are extracted from the synchronous data stream. The instantaneous slope value of each data point in the voltage data segment and the local variance value of each data point in the current data segment at the corresponding time are concatenated according to the corresponding timestamp to generate a time series; The minimum embedding dimension is obtained by calculating the time series using the spurious nearest neighbor method. The optimal delay time is obtained by performing calculations on the time series using the mutual information method. Based on the minimum embedding dimension and the optimal delay time, the time series is reconstructed using delayed coordinate phase space to generate a high-dimensional phase space trajectory.

9. The portable detection method for a dynamically excited miniature oxygen sensor as described in claim 8, characterized in that, The calculation of the maximum Lyapunov exponent of the high-dimensional phase space trajectory specifically includes: The initial phase point in the high-dimensional phase space trajectory is used as the reference phase point; The phase point with the smallest Euclidean distance to the reference phase point in the high-dimensional phase space trajectory is taken as the neighboring phase point; The reference phase point and the neighboring phase point are tracked and calculated along the time axis, and the Euclidean distance between the reference phase point and the neighboring phase point after each time step is calculated as the trajectory separation distance; When the trajectory separation distance is greater than the separation threshold during the tracking calculation process, the current tracking calculation process is taken as the evolution time period, and a new phase point is selected in the high-dimensional phase space trajectory to replace the neighboring phase point. The tracking calculation is then performed based on the replaced neighboring phase point to calculate the next evolution time period until the high-dimensional phase space trajectory has been traversed. For each evolution time period, the initial Euclidean distance between the reference phase point and the neighboring phase point at the start of evolution is obtained, and the final Euclidean distance between the reference phase point and the neighboring phase point at the end of evolution is obtained. Calculate the natural logarithm of the ratio of the terminating Euclidean distance to the initial Euclidean distance; Dividing the natural logarithm by the time step of the corresponding evolution period yields the Euclidean distance divergence rate of the evolution period. Calculate the average of the divergence rates of each Euclidean distance, and use it as the average divergence rate; The average divergence rate is taken as the maximum Lyapunov exponent.

10. A portable detection system for a small oxygen sensor based on dynamic excitation, which is applied in the portable detection method for a small oxygen sensor based on dynamic excitation as described in any one of claims 1-9, characterized in that, It includes a data acquisition module, a calculation module, and a decision-making module; The acquisition module is used to acquire the output voltage signal and pump current signal of the oxygen sensor and generate a synchronous data stream; The calculation module is used to construct a synchronization factor sequence based on the synchronization data stream, input the synchronization factor sequence into the improved Bayesian online change point detection model to calculate the posterior probability value, and output the change point when the posterior probability value is greater than the preset confidence threshold, and use the change point as the extraction starting point. The determination module is used to extract the voltage slope at the end of the response time phase, and extract the pump current data segment with the extraction start time as the starting time. Based on the pump current data segment, the initial pump current fluctuation is calculated. A time series is constructed based on the voltage slope at the end of the response time phase and the initial pump current fluctuation. The time series is mapped to a high-dimensional phase space trajectory through phase space reconstruction. The maximum Lyapunov exponent of the high-dimensional phase space trajectory is calculated to obtain the current parameters of the system under test. When a single parameter is qualified but the maximum Lyapunov exponent exceeds the preset warning threshold, a potential failure is determined.

Images