A pressure sensor fault detection analysis processing system

Abstract

The application discloses a pressure sensor fault detection analysis processing system, comprising: a real-time parameter sequence acquisition module, a local parameter sequence determination module, wherein the local parameter sequence is used for representing an ordered sequence of K adjacent real-time parameters of each real-time parameter; a to-be-detected vector construction module, used for constructing a plurality of local parameter sequences into a to-be-detected vector; a fault information output module, used for inputting the to-be-detected vector into a pre-trained fault detection model to obtain a fault label and a fault level of the pressure sensor; wherein the fault label is node extraction in the forward propagation of the to-be-detected vector, and the fault level is mapping after fitting of the local parameter sequence; the application can represent the evolution trend of potential faults in the running process of the sensor in detail by recognizing the mutation parameter and the neighborhood and constructing the local parameter sequence and comprehensively quantifying the local fault level.

Description

TECHNICAL FIELD

[0001] The application relates to a fault analysis processing system, in particular to a pressure sensor fault detection analysis processing system. BACKGROUND

[0002] The pressure sensor fault detection technology usually relies on the modeling of full parameter sequence, that is, directly based on the full parameter sequence continuously collected by the sensor during operation to identify faults. The patent document with the patent publication number CN116878728B discloses a pressure sensor fault detection analysis processing system, which guarantees the transmission quality of the pressure sensor, prolongs the service life of the pressure sensor to a certain extent, and further provides strong data support for subsequent fault analysis of the pressure sensor.

[0003] However, since the amount of data generated by the sensor during normal operation is large and most of it is stable state data, if the parameters of the sensor during normal operation are used as training samples in the prior art, the effective fault proportion in the training samples will be low, the training efficiency will be low, and false positives and false negatives will be easily caused. In addition, the existing method usually judges faults based on the overall parameter abnormality degree, generally lacks modeling of local abnormal evolution process, and it is difficult to accurately identify potential evolving faults. SUMMARY

[0004] In view of the deficiencies of the prior art, the application provides a pressure sensor fault detection analysis processing system, which identifies mutation parameters and defines local parameter sequences based on the mutation parameters to calculate the fault level of the pressure sensor, so as to solve the technical problems in the background art.

[0005] To achieve the above purpose, the application is implemented by the following technical solutions:

[0006] A real-time parameter sequence acquisition module is configured to acquire N types of real-time parameter sequences in a current time period of the pressure sensor;

[0007] A local parameter sequence determination module is configured to determine a plurality of local parameter sequences for each real-time parameter of the real-time parameter sequence;

[0008] The local parameter sequence is used to represent an ordered sequence of K adjacent real-time parameters of each real-time parameter;

[0009] A to-be-detected vector construction module is configured to construct the plurality of local parameter sequences into a to-be-detected vector;

[0010] A fault information output module is configured to input the to-be-detected vector into a pre-trained fault detection model to obtain a fault label and a fault level of the pressure sensor;

[0011] Wherein, the fault label is node extraction in the forward propagation of the to-be-detected vector, and the fault level is mapping after fitting of the local parameter sequence.

[0012] In some specific embodiments, the modeling step of the fault detection model comprises:

[0013] A1, obtaining a mutation parameter sequence of the pressure sensor in an observation time period;

[0014] A2, extracting sequence macro features from the mutation parameter sequence, and positioning a local parameter sequence based on the sequence macro features;

[0015] A3, constructing a fault sample vector according to the local parameter sequence and calculating a local fault level of the local parameter sequence;

[0016] A4, pairing the fault sample vector with the corresponding local fault level to form a fault detection sample;

[0017] A5, obtaining a plurality of fault detection samples to form a fault detection training set for supervised training;

[0018] A6, inputting the fault detection training set into a supervised model, taking the local fault level as a supervised target, and generating the fault detection model through supervised training.

[0019] In some specific embodiments, obtaining a mutation parameter sequence of the pressure sensor in an observation time period comprises:

[0020] A1-1, obtaining N types of observation parameter sequences of the pressure sensor in the observation time period;

[0021] Each observation parameter sequence represents a time sequence of the same type of parameter in the observation time period, including a pressure sequence, a voltage sequence, an output noise sequence, and an output delay sequence;

[0022] A1-2, extracting sequence macro features of the same type of parameter from the N types of observation parameter sequences;

[0023] A1-3, obtaining N types of reference parameter sequences of the pressure sensor in a reference historical time period;

[0024] A1-4, extracting reference parameter features of the same type of parameter from the N types of reference parameter sequences;

[0025] A1-5, calculating a parameter feature difference of the same type of parameter according to the reference parameter features and the sequence macro features;

[0026] A1-6, if the parameter feature difference exceeds a set threshold, determining that the observation parameter sequence of the same type of parameter is a mutation parameter sequence.

[0027] In some embodiments, the sequence macro-feature is the overall variation feature of the observation parameter sequence in the observation time period;

[0028] When the observation parameter sequence is a pressure sequence, the sequence macro-feature is the pressure variation rate;

[0029] When the observation parameter sequence is a voltage sequence, the sequence macro-feature is the voltage level difference;

[0030] When the observation parameter sequence is an output noise sequence, the sequence macro-feature is the noise energy density;

[0031] When the observation parameter sequence is an output delay sequence, the sequence macro-feature is the average delay time.

[0032] In some embodiments, the A1-1 comprises:

[0033] A1-1-1, anchoring the current observation time point;

[0034] A1-1-2, determining the observation time period by backtracking from the current observation time point for a pre-defined time length;

[0035] A1-1-3, anchoring the type of parameter to be acquired;

[0036] A1-1-4, acquiring the observation parameter scalar of the type of parameter to be acquired at each continuous time point in the observation time period;

[0037] A1-1-5, defining the time series sequence of the observation parameter scalar at each continuous time point as the same type of observation parameter sequence;

[0038] A1-1-6, switching the type of parameter to be acquired, and acquiring the same type of observation parameter sequence after the switching, until N types of observation parameter sequences are acquired.

[0039] In some embodiments, the sequence macro-feature is extracted from the mutation parameter sequence, and the local parameter sequence is located based on the sequence macro-feature, comprising:

[0040] A2-1, calculating the mutation index of each same type of parameter in the mutation parameter sequence;

[0041] A2-2, if the mutation index exceeds a set threshold, marking the corresponding same type of parameter as a mutation parameter;

[0042] A2-3, acquiring the K adjacent parameters before and after the mutation parameter in the corresponding observation parameter sequence;

[0043] A2-4, arranging the mutation parameter and the K adjacent parameters before and after it in time sequence to form a local parameter sequence.

[0044] In some specific embodiments, the mutation index of each parameter of the same type in the mutation parameter sequence is calculated, including:

[0045] A2-1-1. During the observation period, select several local subsequences in the mutation parameter sequence using a sliding window;

[0046] A2-1-2. Calculate the mean and standard deviation of each local subsequence;

[0047] A2-1-3. For parameters of the same type in each local subsequence, calculate their standard score based on the mean and standard deviation.

[0048] The expression obtained from the standard is:

[0049] ;

[0050] in, This represents the t-th parameter of the same type. This represents the mean of parameters of the same type within a local subsequence. This represents the standard deviation of parameters of the same type within a local subsequence. This represents the standard score of the same type of parameter at time point t within the observation period;

[0051] A2-1-4. Compare the standard score of each parameter of the same type with the set threshold; if the standard score is greater than the set threshold, then the parameter of the same type corresponding to the standard score is determined to be a mutation parameter.

[0052] In some specific embodiments, a fault sample vector is constructed based on a local parameter sequence, including:

[0053] A3-1. Characterize the parameters in the local parameter sequence to obtain K+1 local parameter features;

[0054] A3-2. Obtain the parameter types of the mutation parameter sequence and encode the parameter types to form local parameter label features;

[0055] A3-3. Concatenate the K+1 local parameter features with the label features to construct the fault sample vector.

[0056] In some specific embodiments, the steps for constructing the local fault level include:

[0057] B1. Calculate the trend slope, local variance, and maximum offset of abrupt parameter sequences in the local parameter sequence;

[0058] B2. Substitute the trend slope, local variance, and maximum offset of the mutation parameter into the following formula to determine the local fault level of the local parameter sequence.

[0059] ;

[0060] wherein, represents a local fault level, represents the absolute value of the trend slope; V represents the local variance of the local parameter sequence; M represents the maximum offset of the mutation parameter in the local parameter sequence.

[0061] The application provides a pressure sensor fault detection analysis processing system, which has the following beneficial effects:

[0062] The application extracts a mutation parameter sequence, and takes feature difference as a screening basis, so that only parameter fragments that change abnormally significantly in an observation time period are reserved as training sample sources, the density of fault information in the training sample is effectively improved, the interference of irrelevant stable state data on the model training process is reduced, and the training efficiency of the fault detection model is significantly improved.

[0063] The application constructs a local parameter sequence based on a mutation parameter and its neighborhood, and quantifies a local fault level by comprehensively calculating local trend slope, local variance and mutation offset, so that the evolution trend of potential faults in the sensor operation process can be characterized in detail, and compared with a traditional overall abnormality determination method, the method has higher accuracy in early fault detection. BRIEF DESCRIPTION OF DRAWINGS

[0064] Figure 1 FIG. 1 is a structural block diagram of a pressure sensor fault detection analysis processing system according to the application;

[0065] Figure 2 FIG. 2 is a processing flow diagram of a pressure sensor fault detection analysis processing system according to the application;

[0066] Figure 3 FIG. 3 is a determination flow diagram of a mutation parameter sequence according to the application;

[0067] Figure 4 FIG. 4 is a positioning flow diagram of a local parameter sequence according to the application. DETAILED DESCRIPTION

[0068] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the protection scope of the application.

[0069] Embodiment 1: refer to Figures 1 to 2The application provides a pressure sensor fault detection analysis processing system, the processing system comprises:

[0070] A real-time parameter sequence acquisition module is configured to acquire N types of real-time parameter sequences of a pressure sensor in a current time period;

[0071] A local parameter sequence determination module is configured to determine a plurality of local parameter sequences for each real-time parameter in the real-time parameter sequences;

[0072] The local parameter sequences are used to represent an ordered sequence of K adjacent real-time parameters of each real-time parameter;

[0073] A to-be-detected vector construction module is configured to construct the plurality of local parameter sequences into a to-be-detected vector;

[0074] A fault information output module is configured to input the to-be-detected vector into a pre-trained fault detection model to obtain a fault label and a fault level of the pressure sensor;

[0075] The fault label is a node extraction in a forward propagation of the to-be-detected vector, and the fault level is a mapping after fitting of the local parameter sequences.

[0076] In this embodiment, N types of real-time parameter sequences of a target pressure sensor are acquired in a current time period, and each parameter sequence corresponds to a different physical quantity measurement dimension. For any real-time parameter in the real-time parameter sequences, a plurality of local parameter sequences on a time axis are determined to describe parameter variation characteristics in a local region. The plurality of local parameter sequences are arranged according to a set rule to construct a to-be-detected vector in a unified format. Then, the to-be-detected vector is input into a pre-trained fault detection model, and a corresponding fault label and fault level are output, so that fault detection analysis on a current running state of the pressure sensor is realized.

[0077] Embodiment 2: refer to Figures 3 to 4 The technical solution of this embodiment 2 is different from that of embodiment 1 in that a modeling step of the fault detection model described in embodiment 1 is disclosed, and the modeling step comprises:

[0078] A1, acquiring a mutation parameter sequence of the pressure sensor in an observation time period;

[0079] A2, extracting sequence macro features from the mutation parameter sequence, and positioning a local parameter sequence based on the sequence macro features;

[0080] A3, constructing a fault sample vector and calculating a local fault level of the local parameter sequence according to the local parameter sequence;

[0081] A4, pairing the fault sample vector with the corresponding local fault level to form a fault detection sample;

[0082] A5, acquire a plurality of fault detection samples to form a fault detection training set for supervised training;

[0083] A6, input the fault detection training set to a supervised model, take the local fault level as the supervised target, and generate the fault detection model through supervised training.

[0084] In this embodiment, the mutation parameter sequence of the target pressure sensor in the observation time period is extracted as the basic data for fault modeling. Based on the mutation parameter sequence, the sequence macro feature is further extracted to determine the local abnormal trend, and the local parameter sequence is located accordingly. For each local parameter sequence located, a corresponding fault sample vector is constructed, and the local fault level of each local parameter sequence is calculated. The fault sample vector and the local fault level are paired to form a fault detection sample. By accumulating a plurality of fault detection samples, a fault detection training set required for supervised training is constructed. Finally, the fault detection training set is input to a supervised model, and the local fault level is taken as the supervised target. Through supervised training, a fault detection model for fault detection task is generated.

[0085] Exemplarily, the acquisition of the mutation parameter sequence can adopt the following steps, including:

[0086] A1-1, acquire N types of observation parameter sequences of the pressure sensor in the observation time period;

[0087] Each observation parameter sequence represents the time sequence of the same type of parameter in the observation time period, including: pressure sequence, voltage sequence, output noise sequence, output delay sequence;

[0088] A1-2, extract the sequence macro feature of the same type of parameter from the N types of observation parameter sequences;

[0089] A1-3, acquire N types of reference parameter sequences of the pressure sensor in the reference historical time period;

[0090] A1-4, extract the reference parameter feature of the same type of parameter from the N types of reference parameter sequences;

[0091] A1-5, calculate the parameter feature difference of the same type of parameter according to the reference parameter feature and the sequence macro feature;

[0092] A1-6, if the parameter feature difference exceeds the set threshold, determine that the observation parameter sequence of the same type of parameter is a mutation parameter sequence.

[0093] The embodiment extracts sequence macro features of each type of observation parameter sequence in the observation time period to describe the parameter variation characteristics in the current running state. Meanwhile, corresponding reference parameter sequences in a reference historical time period are collected, and historical features are extracted to represent the parameter behavior in the normal state. By comparing the sequence macro features and the historical features, a feature difference value is calculated, and a preset threshold is used to filter out the observation parameter sequences with significant feature differences, which are identified as mutation parameter sequences. By retaining only the mutation parameter sequences as the source of subsequent training samples, the data segments containing abnormal evolution trends can be effectively focused on, and the interference of redundant samples can be avoided.

[0094] Exemplarily, the sequence macro feature is the overall variation characteristic of the observation parameter sequence in the observation time period, which is used to represent the macro variation trend of the parameter sequence in the running process.

[0095] When the observation parameter sequence is a pressure sequence, the sequence macro feature is a pressure change rate.

[0096] When the observation parameter sequence is a voltage sequence, the sequence macro feature is a voltage level difference.

[0097] When the observation parameter sequence is an output noise sequence, the sequence macro feature is a noise energy density.

[0098] When the observation parameter sequence is an output delay sequence, the sequence macro feature is an average delay time.

[0099] The embodiment extracts sequence macro features matched with the variation characteristics of different types of observation parameter sequences to accurately depict the macro variation trend in the running process. For the pressure sequence, the pressure change rate is extracted as the sequence macro feature, which is used to quantify the rising or falling speed of the pressure signal in the observation time period, reflecting the continuous change behavior of the pressure system. For the voltage sequence, the voltage level difference is extracted as the sequence macro feature, which represents the fluctuation range of the voltage supply by measuring the amplitude between the maximum value and the minimum value. For the output noise sequence, the noise energy density is extracted as the sequence macro feature, which is used to reflect the overall amplitude level of the noise signal in the frequency domain, identifying possible abnormal vibration or interference phenomenon in the signal. For the output delay sequence, the average delay time is extracted as the sequence macro feature, which quantifies the variation trend of the system response delay, assisting in judging the response performance change of the control system. By extracting the feature indicators of different physical quantities, the normal running state and abnormal variation trend can be effectively distinguished.

[0100] In the embodiment, the observation parameter sequence needs to define an observation time period, so as to determine the observation parameter sequence; therefore, step A1-1 comprises:

[0101] A1-1-1, anchor the current observation time point;

[0102] A1-1-2, backtrack from the current observation time point for a pre-defined time length to determine the observation time period;

[0103] A1-1-3, anchor the type of parameter to be acquired;

[0104] A1-1-4, acquire the observation parameter scalar of the type of parameter to be acquired at each continuous time point within the observation time period;

[0105] A1-1-5, define the time sequence of the observation parameter scalar at each continuous time point as the same type of observation parameter sequence;

[0106] A1-1-6, switch the type of parameter to be acquired, acquire the same type of observation parameter sequence after switching, and repeat the above extraction and construction process until N types of observation parameter sequences are acquired.

[0107] The embodiment anchors the current observation time point as the reference starting point for determining the observation data range. Then, the complete observation time period is determined by backtracking from the time point for a pre-defined time length, so as to collect time sequence data with continuity. For each type of parameter to be acquired, the observation parameter scalar at each continuous time point within the observation time period is extracted in sequence to form the corresponding time sequence data set. The continuous observation parameter scalar is arranged in time sequence to construct the same type of observation parameter sequence. After the sequence construction of one type of parameter is completed, the next type of parameter to be acquired is switched, and the above extraction and construction process is repeated until the predetermined N types of observation parameter sequences are acquired.

[0108] Exemplarily, the positioning of the local parameter sequence can adopt the following steps:

[0109] A2-1, calculate the mutation index of each same type of parameter in the mutation parameter sequence;

[0110] A2-2, if the mutation index exceeds a set threshold, mark the corresponding same type of parameter as a mutation parameter;

[0111] A2-3, acquire K adjacent parameters before and after the mutation parameter in the corresponding observation parameter sequence;

[0112] A2-4, arrange the mutation parameter and the K adjacent parameters before and after the mutation parameter in time sequence to form a local parameter sequence.

[0113] The embodiment calculates the mutation indicators of each same type parameter in the mutation parameter sequence to measure the abnormal degree of the parameters at each time point relative to the local characteristics. Based on the set threshold condition, the mutation indicators are screened to identify the parameter points with significantly abnormal standard scores, which are marked as mutation parameters. For each mutation parameter, its neighborhood parameters at the K time points before and after the mutation point are further extracted in the corresponding observation parameter sequence to capture the local evolution characteristics around the mutation point. Finally, the mutation parameters and their neighborhood parameters are arranged in time sequence to form a local parameter sequence. The local parameter sequence takes the mutation point as the center to systematically organize the parameter change information in the local area, providing a representative time sequence segment for fault trend modeling.

[0114] In the embodiment, the step A2-1 is used to calculate the mutation indicators, and the calculation steps of the mutation indicators specifically include:

[0115] A2-1-1, selecting several local subsequences in the mutation parameter sequence in a sliding window manner within the observation time period;

[0116] A2-1-2, calculating the mean and standard deviation of each local subsequence;

[0117] A2-1-3, for each same type parameter in each local subsequence, calculating the standard score thereof according to the mean and standard deviation;

[0118] The expression of the standard score is:

[0119] ;

[0120] wherein, denotes the tth same type parameter, denotes the mean of the same type parameters in the local subsequence, denotes the standard deviation of the same type parameters in the local subsequence, denotes the standard score of the same type parameter at the tth time point within the observation time period;

[0121] A2-1-4, comparing the standard score of each same type parameter with a set threshold value; if the standard score is greater than the set threshold value, it is determined that the same type parameter corresponding to the standard score is a mutation parameter.

[0122] The embodiment is aimed at a mutation parameter sequence, adopts a sliding window mode to select a plurality of local subsequences in an observation time period, and captures the change characteristics of the parameters in a short time scale through localization processing. For each local subsequence, the mean and standard deviation thereof are calculated to describe the central tendency and fluctuation level of the parameter values in the local region. Further, for each same type parameter in the local subsequence, a standard score is calculated according to the corresponding local mean and standard deviation to quantify the deviation degree of the parameter relative to the local statistical characteristics. By comparing the standard score with a set threshold, the parameter points with an abnormally large standard score are screened out and determined as mutation parameters. The method can dynamically evaluate the abnormality degree of the observation parameters in the local region, and improve the accuracy of mutation point identification.

[0123] Exemplarily, the construction of the fault sample vector can adopt the following steps:

[0124] A3-1, feature the parameters in the local parameter sequence to obtain K+1 local parameter features;

[0125] A3-2, obtain the parameter types of the mutation parameter sequence, encode the parameter types, and form the label features of the local parameters;

[0126] A3-3, splice the K+1 local parameter features and the label features to construct the fault sample vector;

[0127] The local fault level is used to measure the trend mutation strength of the mutation parameters corresponding to the neighborhood parameters, so as to quantify the potential fault trend of the local parameter sequence.

[0128] The embodiment is aimed at the local parameter sequence positioned through mutation detection, and sequentially extracts features of each parameter point in the sequence. The extracted K+1 local parameter features are used to describe the change trend and fluctuation characteristics in the local region. Further, the parameter type information of the mutation parameters is extracted and converted into label features through a predefined encoding mode to identify the physical source category of the local sample. Subsequently, the K+1 local parameter features and the label features are spliced to construct a fault sample vector in a unified format, ensuring that the sample contains both local change information and type identification information.

[0129] Further, the construction steps of the local fault level include:

[0130] B1, calculating the trend slope, local variance of the local parameter sequence, and the maximum offset of the mutation parameter;

[0131] B2, substituting the trend slope, local variance, and the maximum offset of the mutation parameter into the following formula to determine the local fault level of the local parameter sequence;

[0132] ;

[0133] wherein, denotes the local fault level, denotes the absolute value of the trend slope, used to describe the overall trend intensity of the local parameter sequence; V denotes the local variance of the local parameter sequence, used to describe the overall faultiness of the local parameter sequence; M denotes the maximum offset of the mutation parameter in the local parameter sequence, used to represent the severity of the local mutation fault inside the local parameter sequence.

[0134] The embodiment extracts the mutation parameter and its adjacent parameters before and after the mutation parameter to form a local parameter sequence; then, based on the local parameter sequence, the following three types of sequence local features are calculated respectively:

[0135] Trend slope (S): used to depict the strength of the overall trend of the local parameter sequence, reflecting whether there is a persistent offset or drift phenomenon in the sensor output within the observation time range;

[0136] Local variance (V): used to represent the volatility of the local parameter sequence, reflecting whether there is a fault jitter, dispersion or stability decline behavior in the sensor output;

[0137] Maximum offset (M): used to depict the instantaneous mutation amplitude of the mutation parameter relative to the mean of the adjacent parameters, reflecting the severity of the sudden fault event (such as short circuit, impact, fracture, etc.).

[0138] After obtaining the three types of sequence local features, in order to integrate the contributions of each local feature to the local fault trend, the following fusion strategy is adopted:

[0139] The absolute value of the trend slope and the local variance are added to form the overall faultiness basis quantity; then, taking the maximum offset as the suppression factor, the overall basis quantity is divided by 1+M, wherein the addition of 1 is used to avoid division by zero fault, and at the same time, when the maximum offset is larger, the modulation effect on the overall faultiness is stronger, thereby highlighting the dominant position of the sudden fault; finally, the above ratio is logarithmically transformed, compressing the influence of extreme fault values, improving the sensitivity to subtle change trends, while maintaining the controllable range and numerical stability of the fault level.

[0140] The above-described embodiments can be implemented in part or in whole through software, hardware, firmware or any combination thereof. When implemented in software, the above-described embodiments can be implemented in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When loaded and executed by a computer, the computer instructions or computer programs cause the computer to perform the above-described processes or functions. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable apparatus. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, from a website site, a computer, a server or a data center to another website site, a computer, a server or a data center through a wired (for example, infrared, wireless, microwave, etc.) manner.

[0141] The computer-readable storage medium can be any available medium or a collection of medium accessible by a computer or a data storage device such as a server, a data center, etc. containing one or more available medium. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a magnetic tape), an optical medium (for example, a compact disk, a DVD), or a semiconductor medium. The semiconductor medium can be a solid state disk. DVD

[0142] In several embodiments provided in the present application, it should be understood that the disclosed system, device and method can be implemented in other ways. For example, the above-described device embodiments are merely illustrative, for example, a plurality of units or components can be combined or integrated into another system, or some features can be omitted or not executed. In addition, the coupling or direct coupling or communication connection between the shown or discussed units can be indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0143] The above description is merely a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered within the protection scope of the present application.​

Claims

1. A pressure sensor fault detection analysis processing system, characterized by, The method comprises the following steps: a real-time parameter sequence acquisition module is configured to acquire N types of real-time parameter sequences of the pressure sensor in a current time period; a local parameter sequence determination module is configured to determine a plurality of local parameter sequences for each real-time parameter of the real-time parameter sequence; wherein the local parameter sequence is used to represent an ordered sequence of K adjacent real-time parameters of each real-time parameter; a to-be-detected vector construction module is configured to construct the plurality of local parameter sequences into a to-be-detected vector; a fault information output module is configured to input the to-be-detected vector into a pre-trained fault detection model to obtain a fault label and a fault level of the pressure sensor; wherein the fault label is a node extraction in the forward propagation of the to-be-detected vector, and the fault level is a mapping after fitting of the local parameter sequence; the modeling steps of the fault detection model comprise: acquiring a mutation parameter sequence of the pressure sensor in an observation time period; extracting sequence macro features from the mutation parameter sequence, and positioning local parameter sequences based on the sequence macro features; the step of extracting sequence macro features from the mutation parameter sequence and positioning local parameter sequences based on the sequence macro features comprises: calculating a mutation index of each same-type parameter in the mutation parameter sequence; if the mutation index exceeds a set threshold, marking the corresponding same-type parameter as a mutation parameter; acquiring K adjacent parameters before and after the mutation parameter in the corresponding observation parameter sequence; arranging the mutation parameter and the K adjacent parameters before and after the mutation parameter in time sequence to form a local parameter sequence; the step of calculating the mutation index of each same-type parameter in the mutation parameter sequence comprises: in the observation time period, selecting a plurality of local subsequences in the mutation parameter sequence in a sliding window; calculating the mean and standard deviation of each local subsequence; for each same-type parameter in each local subsequence, calculating a standard score thereof according to the mean and standard deviation; the expression of the standard score is: ； wherein, denotes the t-th same type parameter, denotes the mean of the same type parameters in the local sub-sequence, denotes the standard deviation of the same type parameters in the local sub-sequence, denotes the standard score of the t-th time point of the same type parameter within the observation time period; comparing the standard score of each same-type parameter with a set threshold; if the standard score is greater than the set threshold, determining that the same-type parameter corresponding to the standard score is a mutation parameter.

2. A pressure sensor fault detection analysis processing system according to claim 1, wherein, the modeling steps of the fault detection model further comprise: constructing a fault sample vector and calculating a local fault level of the local parameter sequence according to the local parameter sequence; pairing the fault sample vector with the corresponding local fault level to form a fault detection sample; acquiring a plurality of fault detection samples to form a fault detection training set for supervised training; inputting the fault detection training set into a supervised model, taking the local fault level as a supervised target, and generating the fault detection model through supervised training.

3. A pressure sensor fault detection analysis processing system according to claim 2, wherein, the step of acquiring the mutation parameter sequence of the pressure sensor in the observation time period comprises: acquiring N types of observation parameter sequences of the pressure sensor in the observation time period; each observation parameter sequence represents a time sequence of same-type parameters in the observation time period, including a pressure sequence, a voltage sequence, an output noise sequence, and an output delay sequence; extracting sequence macro features of the same-type parameters from the N types of observation parameter sequences; acquiring N types of reference parameter sequences of the pressure sensor in a reference historical time period; extracting reference parameter features of the same-type parameters from the N types of reference parameter sequences; According to the reference parameter feature and the sequence macro feature, a parameter feature difference of the same type parameter is calculated; If the parameter feature difference exceeds a set threshold, it is determined that the observation parameter sequence of the same type parameter is a mutation parameter sequence.

4. A pressure sensor fault detection analysis processing system according to claim 3, wherein, The sequence macro feature is an overall change feature of the observation parameter sequence in the observation time period; When the observation parameter sequence is a pressure sequence, the sequence macro feature is a pressure change rate; When the observation parameter sequence is a voltage sequence, the sequence macro feature is a voltage level difference; When the observation parameter sequence is an output noise sequence, the sequence macro feature is a noise energy density; When the observation parameter sequence is an output delay sequence, the sequence macro feature is an average delay time.

5. A pressure sensor fault detection analysis processing system according to claim 4, wherein, N types of observation parameter sequences of the pressure sensor in the observation time period are obtained, including: Anchoring a current observation time point; Backtracking from the current observation time point for a pre-defined time length to determine an observation time period; Anchoring a parameter type to be obtained; Obtaining an observation parameter scalar of the parameter type to be obtained at each continuous time point in the observation time period; Defining a time sequence of the observation parameter scalar at each continuous time point as a same type observation parameter sequence; Switching the parameter type to be obtained to obtain a switched same type observation parameter sequence until N types of observation parameter sequences are obtained.

6. A pressure sensor fault detection analysis processing system according to claim 2, wherein, According to the local parameter sequence, a fault sample vector is constructed, including: Characterizing parameters in the local parameter sequence to obtain K+1 local parameter features; Obtaining a parameter type of the mutation parameter sequence and encoding the parameter type to form a label feature of the local parameter; Splicing the K+1 local parameter features and the label feature to construct the fault sample vector.

7. A pressure sensor fault detection analysis processing system according to claim 6, wherein, The construction step of the local fault level includes: B1, calculating a trend slope, a local variance and a maximum offset of the mutation parameter of the local parameter sequence; B2, substituting the trend slope, the local variance and the maximum offset of the mutation parameter into the following formula to determine the local fault level of the local parameter sequence. ； wherein, denotes a local fault level, denotes the absolute value of the trend slope; V denotes the local variance of the local parameter sequence; M denotes the maximum deviation of the mutation parameters in the local parameter sequence.

Images