Abnormal operating condition monitoring method for oil pump motor and its outlet pipeline based on sound recognition

Abstract

A method for monitoring abnormal operating conditions of oil pump motors and their outlet pipelines based on sound recognition applies voiceprint recognition technology to the speed-regulating hydraulic system of a hydropower station. Through the rational layout of on-site voiceprint sensors, real-time monitoring of the operating conditions of the oil pump motor and its outlet pipeline is achieved, improving the level of intelligent equipment management and serving as an industry benchmark. By combining a voiceprint fault sample library with deep learning analysis of multi-dimensional factors such as pressure, liquid level, and temperature, a multi-dimensional logical judgment process based on voiceprint recognition was designed to identify abnormal operating conditions of the oil pump motor and its outlet pipeline. This method enables real-time monitoring and rapid location of abnormal operating conditions such as base vibration, loss of lubricating oil in the oil pump motor, broken pump blades, check valve malfunction, leaking or leaking drain valves, frequent loading and unloading of the oil pump, and twitching of the relay pipeline. This allows for timely response measures such as automatic switching between the main and standby oil pump motor units or automatic load adjustment of the units. Using voiceprint sensors and voiceprint recognition technology, on-site equipment status is monitored in real time, eliminating the uncertainty associated with manual inspections and improving the efficiency of fault diagnosis and analysis.

Description

Technical Field

[0001] The present invention belongs to the technical field of abnormal operating condition monitoring of hydropower stations, and in particular relates to a method for monitoring abnormal operating conditions of an oil pump motor and its outlet pipeline based on sound recognition. Background Art

[0002] The oil pump motor is a critical driving device in a hydropower station's speed-regulating hydraulic system, along with its associated ancillary equipment such as the base, valves, and pipelines. Failures in the oil pump motor and its outlet pipeline can cause malfunctions in the speed-regulating hydraulic system, disrupting the unit's normal standby function and, in severe cases, forcing the unit to shut down. Therefore, real-time monitoring of the operating conditions of the oil pump motor and its outlet pipeline is essential to facilitate timely detection and resolution.

[0003] The current monitoring methods for the oil pump motor and its outlet pipeline are:

[0004] 1. Through periodic on-site inspections by personnel, abnormal noises during the operation of the oil pump motor and motor reversal can be detected to determine faults such as lack of motor lubricating oil, failure of pipeline check valves, and vibration of the oil pump motor base.

[0005] 2. By recording the oil level in the oil tank, manually analyze the frequency of loading and unloading of the oil pump motor, compare it with the experience value, determine whether the loading and unloading is frequent, and check on site whether the pipeline is twitching.

[0006] 3. Check the oil pump motor and its accessories for oil leakage through on-site inspection or video monitoring.

[0007] The disadvantages of the prior art are:

[0008] 1. Due to the different sensitivity of the human body to sound, the judgment of whether the sound of the oil pump motor is abnormal will also vary from person to person. It is inefficient and lacks real-time performance. There are many reasons for the abnormality, such as lack of lubricating oil, broken pump blades, base vibration, etc. The cause of the abnormality cannot be accurately located.

[0009] 2. Whether the check valve function is invalid needs to be indirectly judged by manually checking whether the oil pump motor is reversed. The rotating parts of the oil pump motor are all covered with protective covers, and it is difficult to manually check the reversal situation. If you look closely at the rotating parts, there are safety hazards.

[0010] 3. Image monitoring can only detect oil leakage or oil spraying on the peripheral part of the equipment. It cannot identify internal leakage caused by valves not being closed tightly or failure of internal seals of the equipment. Manual judgment also requires a long time to analyze a large amount of data, which is inefficient and lacks real-time performance.

[0011] 4. Manual analysis of the frequency of oil pump loading and unloading has problems such as heavy workload and time lag, and the determination of empirical values ​​is not uniform. Frequent loading and unloading will cause pipeline twitching, and on-site inspection is required.

[0012] To sum up, existing technologies mainly rely on manual periodic on-site inspections, which are inefficient, lack reliability and real-time performance, and also pose certain safety risks. Image monitoring cannot identify internal leakage caused by valves not being closed tightly and failure of internal seals of equipment, making manual judgment difficult. The diagnosis of initial faults often lags behind the occurrence of secondary faults. Summary of the Invention

[0013] The technical problem to be solved by the present invention is to provide a method for monitoring the abnormal working conditions of an oil pump motor and its outlet pipeline based on sound recognition. The method uses a voiceprint sensor and voiceprint recognition technology to monitor the status of on-site equipment in real time, eliminates the uncertainty factors of manual inspection, and improves the efficiency of fault diagnosis and analysis.

[0014] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0015] A method for monitoring abnormal working conditions of an oil pump motor and its outlet pipeline based on sound recognition, comprising the following steps:

[0016] Step 1: Collect voiceprint information:

[0017] Step 1.1. Arrange soundprint sensors on the pump body, motor, base, and valve outlet pipe of the oil pump motor; arrange temperature measuring resistors on the motor and oil pump bearings; arrange pressure sensors at the oil pump outlet, oil pressure tank, and pressure regulating valve inlet; arrange liquid level sensors on the oil sump and oil pressure tank; use the soundprint sensors to collect soundprint information, use the temperature measuring resistors to collect temperature information, use the pressure sensors to collect pressure information, and use the liquid level sensors to collect liquid level information;

[0018] Step 1.2: The voiceprint information is processed by the front-end box and then transmitted to the edge computing gateway for preliminary screening. It is then transmitted to the server through the switch for data analysis.

[0019] The temperature information, pressure information and liquid level information are sent to the local LCU unit through the remote I / O module, and then transmitted to the server through the switch for comprehensive data analysis;

[0020] Step 2: Extract voiceprint features from the voiceprint information and process the voiceprint features:

[0021] Mel-frequency cepstral coefficients (MFCCs) are used to extract voiceprint information. The UBM method and the i-vector method are used to convert voiceprint features into model vectors that can be used for calculation and analysis. The probabilistic linear discriminant analysis (PLDA) method is used to compare and score the voiceprint features.

[0022] Step 3: Perform comprehensive logical judgment on the four analysis results;

[0023] Step 4: Output the final fault result.

[0024] Preferably, the MFCC extraction method process in step 2 is as follows:

[0025] 1) The collected voiceprint information is pre-emphasized through a high-pass filter and then divided into overlapping frames. A Hamming window function is applied to each frame to offset the assumption of infinite data made by the fast Fourier transform and reduce spectral leakage;

[0026] 2) Perform a fast Fourier transform on each frame and calculate the power spectrum. The spectrum is extracted using a Mel filter bank. The spectrum is logarithmically calculated to obtain the energy. Then, a discrete cosine transform is applied to decorrelate the spectrum.

[0027] 3) By extracting dynamic differential parameters, the MFCC feature vector consisting of N-dimensional MFCC parameters + frame energy is finally obtained; the dynamic differential parameters include first-order differential and second-order differential.

[0028] Preferably, the UBM calculation method process in step 2 is as follows:

[0029] First, a large amount of non-fault state voiceprint information is collected to train a UBM. Then, a small amount of fault state voiceprint information is used to adjust the UBM parameters through an adaptive algorithm to obtain the target model parameters.

[0030] Preferably, in step 2, the i-vector analysis method process is as follows:

[0031] i-vector defines a low-dimensional vector To represent an audio segment;

[0032] ;

[0033] M is the ideal feature supervector corresponding to the device being modeled, multiplied by the i-vector by the T matrix, and added to the UBM mean supervector;

[0034] m is the UBM mean supervector. Assuming that the UBM contains C Gaussian mixture components g, then m is the mean vector m of all mixture components. c ,c=1,...,C, the dimension of m is C*F, the dimension of T is C*F×R, and F represents the dimension of MFCC features;

[0035] For each mixture component c, parameter mixture weight w c , and covariance matrix Σc; further split the T matrix into a combination of C Vc, the dimension of Vc is F*R,

[0036] ;

[0037] Define a piece of audio feature data X, whose feature dimension is F and time sequence is T, that is, X = X1, . . . , X T ; The subset of X that belongs to the cth Gaussian component is X c , a certain frame X in the subset t c ,but:

[0038] ;

[0039] This gives the following formula:

[0040] ;

[0041] The calculation formula of i-vector is:

[0042] ;

[0043] Among them, N(u) dimension is a diagonal matrix of CF*CF, and the diagonal block is NcI, (c=1,...,C), is a CF×1 supervector, consisting of all first-order BW statistics composition.

[0044] Preferably, in step 2, the PLDA analysis method process is as follows:

[0045] Assume that the training data audio consists of audio from I devices, where each device has J different audio segments; the i-vector of the j-th audio segment of the i-th device is denoted as D ij , then PLDA defines:

[0046] ;

[0047] Among them, the device information part is , is only related to device i and describes the differences between devices;

[0048] The noise part is ,Describe the differences between devices;

[0049] represents the mean of all training data;

[0050] F is considered as the identity space, which contains information that can be used to represent various devices;

[0051] Think of it as the identity of a specific device or the location of the device in the identity space;

[0052] G is considered as the error space, which contains information that can be used to represent different audio variations of the same device;

[0053] It represents the position in G space;

[0054] is the final residual noise term, representing something that has not yet been explained;

[0055] After PLDA model training, the likelihood ratio strategy is used for device identification as follows:

[0056] ;

[0057] As in the above formula, x i and x p They are the i-vector vectors of two audios. The assumption that these two audios come from the same space is M1, and the assumption that they come from different spaces is M0. is the likelihood function that the two audios come from the same space; , x i and x p Likelihood functions from different spaces; by calculating the log-likelihood ratio, the similarity of two audios is calculated and scored;

[0058] The higher the ratio, the higher the score, and the more likely it is that the two audios belong to the same device;

[0059] The lower the ratio, the lower the score, and the less likely it is that the two audios belong to the same device.

[0060] Preferably, the sub-steps in step 3 are as follows:

[0061] Through PLDA model analysis, a voiceprint feature fault sample library for the corresponding equipment is established. If the voiceprint feature data collected on-site successfully matches the fault sample library, the fault type is directly output; the voiceprint data that does not match is scored by the PLDA model. If the score is abnormal, a fault judgment is made based on multiple factors including pressure, liquid level and temperature. The fault sample is manually confirmed and marked, and the fault sample library is supplemented. The steps are as follows:

[0062] ①The motor lubricating oil is missing:

[0063] Motor soundprint data S 1-1 or S 1-2 If the score is abnormal, analyze the rising trend of the motor bearing temperature T1 to =1min is the cycle, the current time t is the starting point, calculate the rising rate of T1 , the formula is as follows:

[0064] ;

[0065] If the rising rate of T1 in the last three cycles 、 、 The relationship between the three is: < , it can be judged that the motor bearing temperature rises abnormally, and the comprehensive voiceprint data score is abnormal, which can be determined to be the lack of motor lubricating oil;

[0066] ②The oil pump is missing lubricating oil:

[0067] Oil pump voiceprint data S 2-1 or S 2-2 If the score is abnormal, analyze the rising trend of the oil pump bearing temperature T2. If the rising rate of T2 in the last three cycles meets < , it can be determined that the oil pump bearing temperature rises abnormally, and the comprehensive soundprint data score is abnormal, which can be determined as the oil pump lubricating oil is missing;

[0068] ③ Abnormal base vibration:

[0069] If the soundprint data S3 score on the upper surface of the base is abnormal, and the motor and oil pump lubricant loss features do not match, then the base vibration is determined to be abnormal;

[0070] ④Pump blade fracture:

[0071] Oil pump voiceprint data S 2-1 or S 2-2 The score is abnormal, the rising trend of the oil pump bearing temperature T2 is normal, and the oil pump outlet pressure P1 is lower than the normal loading and unloading pressure under loading and unloading conditions, that is, P 1加 P 加 And P 1卸 P 卸 , it can be determined that the oil pump efficiency is abnormal. Combined with the normal motor sound pattern data, it can be determined that the oil pump blade is broken;

[0072] ⑤ Check valve function failure:

[0073] When the oil pump motor stops working, the oil pump outlet pipe soundprint data S5 score is abnormal, and P1 0, and the soundprint data score of the oil pump motor is abnormal, it can be determined that the oil pump outlet check valve function is malfunctioning, causing the oil pump motor to reverse;

[0074] ⑥ Frequent loading and unloading of the oil pump:

[0075] The S6 score of the voiceprint data of the loading and unloading valve group is abnormal. Analyze the time interval between loading and unloading of the oil pump. , when the oil tank pressure is P3 When the pressure reaches 6.1 MPa, the oil pump starts loading and the time is recorded as , next time P3 When the pressure reaches 6.1 MPa, the time is recorded as ,but = - ,like ,in If it is the set value of loading and unloading interval, it is determined that the oil pump is loaded and unloaded frequently;

[0076] ⑦ The oil drain valve is not closed tightly or leaks internally:

[0077] The S4 score of the oil tank drain valve voiceprint data is abnormal, and the time it takes for the oil pump to load once Too long, that is > , is the setting value of the loading time, and the frequent loading and unloading characteristics have been matched. The total oil volume V of the hydraulic system is calculated by combining the liquid levels H1 and H2 of the pressure oil tank and the oil collection tank. If V=V0, where V0 is the initial total oil volume of the hydraulic system, it is determined that the oil drain valve is not closed tightly or there is internal leakage;

[0078] ⑧Servomotor pipeline twitching:

[0079] If the scores of the soundprint data S7 of the pressure-regulating valve body and the soundprint data S8 or S9 of the pressure-regulating valve outlet pipeline are abnormal, the frequent loading and unloading characteristics of the oil pump have been matched, and the total oil volume V=V0, it can be determined that the relay pipeline is twitching.

[0080] Preferably, after the above faults are matched or confirmed, an alarm signal and a fault message are output through human-computer interaction of the upper computer. After any one of faults ①-⑤ is confirmed to have occurred, the main-standby automatic switching function of the oil pump motor group should be activated, the standby oil pump motor group should be started, and the faulty group should be deactivated and marked as unavailable for maintenance;

[0081] After fault ⑥ is confirmed to have occurred, only the alarm signal is output;

[0082] If faults ⑥ and ⑦ occur simultaneously, it is considered that the oil drain valve is not closed tightly or is leaking internally, and an alarm signal is output, and the startup conditions are locked;

[0083] Faults ⑥ and ⑧ occurred simultaneously, and were confirmed to be caused by twitching of the relay pipeline. An alarm signal was output, and the AGC was activated to adjust the unit output. According to the actual output of the unit, the output was increased or decreased in steps of 10MW to adjust the unit operating conditions. This process requires manual monitoring, and manual intervention and adjustment and on-site confirmation when necessary.

[0084] Preferably, in step 1, the voiceprint sensor adopts a patch-type bone conduction pickup, which is arranged on: two sides of the motor housing, two sides of the pump housing, the upper surface of the base, the oil pump outlet pipeline, the loading and unloading valve group, the oil tank drain valve, the pressure regulating valve body and the pressure regulating valve outlet pipeline;

[0085] The temperature measuring resistors are arranged at: motor bearings and oil pump bearings;

[0086] Pressure sensor layout: oil pump outlet, oil pressure tank and pressure regulating valve inlet;

[0087] Liquid level sensor arrangement: oil sump and oil pressure tank.

[0088] The present invention can achieve the following beneficial effects:

[0089] 1. Using voiceprint sensors and voiceprint recognition technology to monitor the status of on-site equipment in real time, eliminating the uncertainty of manual inspection and improving the efficiency of fault diagnosis and analysis.

[0090] 2. Establish a voiceprint fault sample library for oil pump motors and outlet pipelines, which is universal in the industry and has great reference value.

[0091] 3. The voiceprint recognition technology is innovatively applied to the speed regulating hydraulic system of the hydropower station. Through the rational layout of the on-site voiceprint sensors, the operating conditions of the oil pump motor and its outlet pipeline are monitored in real time, which improves the intelligent level of equipment management and serves as an industry model.

[0092] 4. Through the voiceprint fault sample library combined with deep learning analysis of multi-dimensional factors such as pressure, liquid level, and temperature, a logical judgment process of multi-dimensional factors of abnormal operating conditions of the oil pump motor and outlet pipeline based on voiceprint recognition is designed. It can realize real-time monitoring and rapid positioning of abnormal operating conditions such as base vibration, lack of lubricating oil in the oil pump motor, broken pump blades, failure of check valve function, oil drain valve not closed tightly or internal leakage, frequent loading and unloading of the oil pump, and twitching of the relay pipeline, and timely take countermeasures such as automatic switching of the main and standby oil pump motor groups or automatic adjustment of the unit load.

[0093] 5. Through the Mel-frequency cepstral coefficient MFCC algorithm, the universal background model UBM method and the i-vector method for calculating identity vector features, the voiceprint information of the oil pump motor and outlet pipeline equipment is converted into a model vector that can be used for calculation and analysis, making the recognition and analysis of voiceprint information more accurate and reasonable, and the judgment of equipment failure more reliable. BRIEF DESCRIPTION OF THE DRAWINGS

[0094] The present invention will be further described below with reference to the accompanying drawings and examples:

[0095] Figure 1 Flowchart of the present invention;

[0096] Figure 2 This is a layout diagram of the voiceprint sensor, temperature resistor, pressure sensor, and liquid level sensor of the present invention;

[0097] Figure 3 This is a data transmission diagram of the present invention;

[0098] Figure 4 Figure 1 is a diagram of the voiceprint feature extraction and processing method of the present invention;

[0099] Figure 5 Extract 13th-order MFCC feature 3D graph for 500ms voiceprint data of the present invention;

[0100] Figure 6 This is a flow chart of the comprehensive evaluation of motor fault soundprints of the present invention;

[0101] Figure 7 This is a flow chart of the comprehensive evaluation of the oil pump fault soundprint of the present invention;

[0102] Figure 8 This is a flow chart of comprehensive evaluation of sound patterns of check valve failures according to the present invention;

[0103] Figure 9 This is a flow chart of the comprehensive evaluation of frequent sound patterns of oil pump loading and unloading according to the present invention;

[0104] Figure 10 This is a flow chart for comprehensive evaluation of the internal leakage sound pattern of the oil discharge valve of the present invention;

[0105] Figure 11 This is a flow chart of the comprehensive evaluation of the twitching soundprint of the servomotor pipeline of the present invention;

[0106] Figure 12 This is the fault result output and preventive control measures logic diagram of the present invention.

[0107] Figure 2 Middle: S 1-1 / S 1-2 、S 2-1 / S 2-2 , S3-S9 are voiceprint sensors (S 1-1 、S 1-2 The front and rear are symmetrically arranged. 2-1 、S 2-2 P1-P3 are pressure sensors; H1 / H2 are liquid level sensors; T1 is the motor bearing temperature resistor; T2 is the oil pump bearing temperature resistor. DETAILED DESCRIPTION

[0108] The present invention uses voiceprint sensors and voiceprint recognition technology to conduct real-time status monitoring of the oil pump motor and outlet pipeline of the speed regulating hydraulic system of the hydropower station, replacing manual regular on-site inspections, eliminating the lag of manual inspections and the unreliable factors in judging equipment faults; using MFCC, PLDA and other voiceprint recognition technologies to process the collected voiceprint information, and after repeated training, establish a corresponding equipment voiceprint fault sample library, which can be widely used in the judgment of voiceprint fault characteristics of related equipment in the industry; matching the collected voiceprint feature data with the fault sample library, making the judgment accurate and rapid; performing PLDA on the unmatched voiceprint data The model is scored, and those with abnormal scores are judged by integrating multiple factors such as pressure, liquid level, temperature, etc. The fault samples are manually confirmed and marked, and the fault sample library is supplemented. A front-end storage device is arranged in the voiceprint collection section, and the original sound information can be accessed at the front end and can also be read offline, which greatly improves the storage capacity of the terminal device. The real-time fault information that has been matched or marked will send an alarm message in the terminal human-computer interaction and be viewed in the form of a fault message, so as to achieve real-time monitoring of on-site equipment, timely and accurate alarm prompts for abnormal situations, and automatically take countermeasures to eliminate equipment abnormalities in the early stages and avoid serious accidents.

[0109] This invention innovatively applies voiceprint recognition technology to the speed regulation hydraulic system of a hydropower station. Through the rational layout of on-site voiceprint sensors, it achieves real-time monitoring of the operating conditions of the oil pump motor and its outlet pipeline, improves the intelligent level of equipment management, and serves as a model for the industry.

[0110] By combining the voiceprint fault sample library with deep learning analysis of multi-dimensional factors such as pressure, liquid level, and temperature, a multi-dimensional logical judgment process for abnormal operating conditions of the oil pump motor and outlet pipeline based on voiceprint recognition was designed. This allows real-time monitoring and rapid positioning of abnormal operating conditions such as base vibration, lack of lubricating oil in the oil pump motor, broken pump blades, malfunctioning check valves, leaking drain valves, frequent loading and unloading of the oil pump, and twitching of the relay pipeline. This allows for timely response measures such as automatic switching of the main and standby oil pump motor groups or automatic adjustment of the unit load. The specific plan is as follows:

[0111] The preferred solution is Figures 1 to 12 As shown in FIG, a method for monitoring abnormal working conditions of an oil pump motor and its outlet pipeline based on sound recognition mainly comprises the steps of information collection and transmission, voiceprint feature processing, comprehensive logic judgment, and fault result output. Figure 1 The on-site equipment information is collected and transmitted to the host computer monitoring system in real time, and the host computer automatically completes the voiceprint feature processing, comprehensive logic judgment, and fault result output. The specific steps are as follows:

[0112] Step 1: Information collection and transmission:

[0113] Use sensor equipment to collect information: arrange voiceprint sensors, temperature measuring resistors, pressure sensors, liquid level sensors, etc. on the pump body, motor, base and valve outlet pipeline of the oil pump motor respectively. Among them, the voiceprint sensor adopts patch-type bone conduction pickup, which is arranged at: one on each side of the motor casing (symmetrical arrangement), one on each side of the pump casing (symmetrical arrangement), one on the upper surface of the base, one on the oil pump outlet pipeline (near the check valve), one on the loading and unloading valve group (at the outlet pipeline), one on the oil pressure tank drain valve (at the outlet pipeline), one on the pressure regulating valve body, and one on the pressure regulating valve outlet pipeline; temperature measuring resistor arrangement: one on the motor bearing and one on the oil pump bearing; pressure sensor arrangement: one on the oil pump outlet, one on the oil pressure tank, and one on the pressure regulating valve inlet; liquid level sensor arrangement: one on the oil sump and one on the oil pressure tank. The specific locations are as follows: Figure 2 .

[0114] Use computers for data transmission: The collected voiceprint information is first processed by the front-end box, and then screened at the edge computing gateway, and then transmitted to the server through the switch for data analysis. Human-computer interaction can read and write data through the server. The front-end box has storage and voiceprint initial processing functions, and can store original sound information in the front-end box, which is convenient for reading historical data and saving storage space for terminal devices. Temperature, pressure, liquid level and other information are sent to the local LCU unit through the remote I / O module, and transmitted to the server through the switch for comprehensive data analysis, such as Figure 3 .

[0115] Step 2: Use a computer to process voiceprint features:

[0116] Voiceprint feature processing algorithms include: voiceprint MFCC extraction, UBM calculation, i-vector calculation, PLDA analysis, such as Figure 4 .

[0117] 1) MFCC feature extraction (this process requires the use of computer programs):

[0118] MFCC feature extraction includes preprocessing (pre-emphasis, framing, and windowing), fast Fourier transform (FFT), Mel filter bank, logarithmic operation, discrete cosine transform (DCT), and dynamic feature extraction. The collected voiceprint information is pre-emphasized through a high-pass filter and then segmented into (overlapping) frames. A Hamming window function is applied to each frame to offset the FFT's assumption of infinite data size and reduce spectral leakage. Each frame is then subjected to a fast Fourier transform (FFT) and the power spectrum is calculated. The spectrum is extracted through a Mel filter bank, and the logarithmic energy is calculated. The discrete cosine transform (DCT) is then applied for decorrelation. Finally, dynamic differential parameters (including first-order and second-order differentials) are extracted to obtain an MFCC feature vector consisting of N-dimensional MFCC parameters (N / 3 MFCC coefficients + N / 3 first-order differential parameters + N / 3 second-order differential parameters) + frame energy.

[0119] Taking 500 milliseconds of voiceprint data as an example, 0.97 is used as the pre-emphasis coefficient to pre-emphasize the intercepted voiceprint data, and then divided into multiple frames, each 32 milliseconds, with a frame shift of 16 milliseconds (32ms frame data length, the number of samples is the sampling frequency multiplied by time: 16000 * 0.032 = 512; 16ms frame shift length, the number of samples is the sampling frequency multiplied by time: 16000 * 0.016 = 256). Set the number of Mel filters to 24, extract the MFCC features of 500 milliseconds of voiceprint data (frame length 512; frame shift 256; number of MFCC coefficients 24), and draw a 3D graph of the 13th-order MFCC, as shown below. Figure 5 .

[0120] 2) Calculate the UBM (this process requires the use of a computer program):

[0121] The Universal Background Model (UBM) first collects a large amount of voiceprint information from non-fault conditions and trains a UBM. UBM training is essentially the same as Gaussian Mixture Model (GMM) training. The specific method is to iteratively optimize the MFCC parameters of the voiceprint features from non-fault conditions using the Expectation Maximization (EM) algorithm to obtain the UBM model parameters. Then, using a small amount of faulty voiceprint data, the UBM parameters are adjusted using the adaptive MAP algorithm to obtain the target model parameters. Generally, the only parameter adjusted by the adaptive algorithm is the mean m. This approach reduces the number of faulty samples and parameters required for training, facilitating rapid training convergence and decoding calculations on mobile terminals.

[0122] 3) Calculate i-vector:

[0123] i-vector defines a low-dimensional vector To represent an audio segment.

[0124] ;

[0125] M is the ideal feature supervector corresponding to a device being modeled, multiplied by the i-vector by the T matrix, and added to the UBM mean supervector; m is the UBM mean supervector. Assuming that the UBM contains C Gaussian mixture components g, then m is the mean vector m of all mixture components. c ,c=1,...,C, the dimension of m is C*F, the dimension of T is C*F×R, and F represents the dimension of MFCC features.

[0126] For each mixture component c, there is also a parameter mixture weight w c , and the covariance matrix Σc. Further split the T matrix into a combination of C Vc, the dimension of Vc is F*R,

[0127] ;

[0128] Define a piece of audio feature data X, whose feature dimension is F and time sequence is T, that is, X = X1, . . . , X T The subset of X that belongs to the cth Gaussian component is X c , a certain Frame X in the subset t c ,but

[0129] ;

[0130] From this we can get the following formula

[0131] ;

[0132] i-vector calculation formula

[0133] ;

[0134] Among them, N(u) dimension is a diagonal matrix of CF*CF, and the diagonal block is NcI, (c=1,...,C), is a CF×1 supervector, consisting of all first-order BW statistics composition.

[0135] 4. PLDA analysis (this process requires the use of computer programs):

[0136] Probabilistic Linear Discriminant Analysis (PLDA) is also a channel compensation algorithm, also known as the probabilistic LDA algorithm. PLDA is also typically based on i-vector features, providing channel compensation. We assume that the training data audio consists of audio from I devices, where each device has J distinct audio segments. The i-vector of the j-th audio segment of the i-th device is denoted by D ij , then PLDA defines

[0137] ;

[0138] Among them, the device information part is , is only related to device i and describes the differences between devices;

[0139] The noise part is , describing the differences between devices internally.

[0140] represents the mean of all training data;

[0141] F can be regarded as an identity space, which contains information that can be used to represent various devices;

[0142] It can be regarded as the identity of a specific device (or the location of the device in the identity space);

[0143] G can be thought of as an error space, containing information that can be used to represent different audio variations of the same device;

[0144] It represents the position in G space;

[0145] is the final residual noise term, representing what has not yet been explained.

[0146] After PLDA model training, the likelihood ratio strategy is used for device identification as follows:

[0147] ;

[0148] As in the above formula, x i and x p They are the i-vector vectors of two audios. The assumption that these two audios come from the same space is M1, and the assumption that they come from different spaces is M0. is the likelihood function that the two audios come from the same space;

[0149] , x i and x pThe likelihood functions of different sub-spaces are used to calculate the similarity between two audios.

[0150] The higher the ratio, the higher the score, and the more likely it is that the two audios belong to the same device;

[0151] The lower the ratio, the lower the score, and the less likely it is that the two audios belong to the same device.

[0152] Step 3: Comprehensive logical judgment (this process needs to be performed using a computer program):

[0153] Through PLDA model analysis, a voiceprint feature fault sample library for the corresponding equipment is established. If the voiceprint feature data collected on-site successfully matches the fault sample library, the fault type is directly output; the voiceprint data that fails to match is scored by the PLDA model. For those with abnormal scores, the fault is determined based on multiple factors such as pressure, liquid level, and temperature. The fault samples are manually confirmed and marked, and the fault sample library is supplemented. The details are as follows:

[0154] ①The motor lubricating oil is missing:

[0155] Motor soundprint data S 1-1 or S 1-2 If the score is abnormal, analyze the rising trend of the motor bearing temperature T1 to =1min is the cycle, the current time t is the starting point, calculate the rising rate of T1 , the formula is as follows:

[0156]

[0157] If the rising rate of T1 in the last three cycles 、 、 The relationship between the three is: < , it can be judged that the motor bearing temperature rises abnormally, and the comprehensive soundprint data score is abnormal, which can be judged as the lack of motor lubricating oil, such as Figure 6 .

[0158] ②The oil pump is missing lubricating oil:

[0159] Oil pump voiceprint data S 2-1 or S 2-2 If the score is abnormal, analyze the rising trend of the oil pump bearing temperature T2 (calculation method is the same as above). If the rising rate of T2 in the last three cycles meets < , it can be determined that the oil pump bearing temperature rise is abnormal, and the comprehensive soundprint data score is abnormal, which can be determined as the oil pump lubricating oil is missing, such as Figure 7 .

[0160] ③ Abnormal base vibration:

[0161] If the soundprint data S3 score on the upper surface of the base is abnormal and the lubricating oil missing characteristics of the motor and oil pump do not match, it is determined that the base vibration is abnormal.

[0162] ④Pump blade fracture:

[0163] Oil pump voiceprint data S 2-1 or S 2-2 The score is abnormal, the rising trend of the oil pump bearing temperature T2 is normal, and the oil pump outlet pressure P1 is lower than the normal loading and unloading pressure under loading and unloading conditions, that is, P 1加 P 加 And P 1卸 P 卸 , it can be determined that the oil pump efficiency is abnormal. Combined with the normal motor sound pattern data, it can be determined that the oil pump blade is broken. Figure 7 .

[0164] ⑤ Check valve function failure:

[0165] When the oil pump motor stops working, the soundprint data S5 score of the oil pump outlet pipeline (near the check valve) is abnormal, and P1 0, and the soundprint data score of the oil pump motor is abnormal, it can be determined that the oil pump outlet check valve function is malfunctioning, causing the oil pump motor to reverse, such as Figure 8 .

[0166] ⑥ Frequent loading and unloading of the oil pump:

[0167] The S6 score of the voiceprint data of the loading and unloading valve group (at the outlet pipe) is abnormal. Analyze the time interval between loading and unloading of the oil pump. , when the oil tank pressure is P3 When the oil pump starts to load at 6.1MPa, it is recorded as time , next time P3 When the pressure reaches 6.1 MPa, the time is recorded as ,but = - ,like ( is the loading and unloading interval setting value), it is determined that the oil pump is frequently loaded and unloaded. Figure 9 .

[0168] ⑦ The oil drain valve is not closed tightly or leaks internally:

[0169] The S4 score of the voiceprint data of the oil tank discharge valve (outlet pipe) is abnormal, and the time it takes for the oil pump to load once is (The time it takes for P3 to rise from 6.1MPa to 6.3MPa) is too long, i.e. > ( is the setting value of loading time), the loading and unloading frequent characteristics have been matched, and the total oil volume V of the hydraulic system is calculated based on the liquid levels H1 and H2 of the pressure oil tank and the oil collecting tank. If V=V0 (V0 is the initial total oil volume of the hydraulic system), it is determined that the oil drain valve is not closed tightly or there is internal leakage. Figure 10 .

[0170] ⑧Servomotor pipeline twitching:

[0171] If the soundprint data S7 of the pressure distribution valve body and the soundprint data S8 or S9 of the pressure distribution valve outlet pipe are abnormal, the frequent loading and unloading characteristics of the oil pump have been matched, and the total oil volume V=V0, it can be determined that the relay pipe is twitching. Figure 11 .

[0172] Step 4: Fault result output (this process requires the use of a computer program):

[0173] After the above faults are matched or confirmed, the host computer will output alarm signals and fault messages through human-computer interaction, and start corresponding preventive measures, such as Figure 12 .

[0174] After any of the faults ①-⑤ is confirmed to have occurred, the main and standby automatic switching function of the oil pump motor group should be activated, the standby oil pump motor group should be started, and the faulty group should be deactivated and marked as unavailable for maintenance;

[0175] After fault ⑥ is confirmed to have occurred, only the alarm signal is output;

[0176] If faults ⑥ and ⑦ occur simultaneously, it is considered that the oil drain valve is not closed tightly or is leaking internally, and an alarm signal is output, and the startup conditions are locked;

[0177] Faults ⑥ and ⑧ occurred simultaneously, and were confirmed to be caused by twitching of the relay pipeline. An alarm signal was output, and the AGC was activated to adjust the unit output. According to the actual output of the unit, the output was increased or decreased in steps of 10MW to adjust the unit operating conditions. This process requires manual monitoring, and manual intervention and adjustment and on-site confirmation when necessary.

[0178] The above embodiments are merely preferred technical solutions of the present invention and should not be construed as limiting the present invention. The scope of protection of the present invention shall be the technical solutions set forth in the claims, including equivalent alternatives to the technical features of the technical solutions set forth in the claims. In other words, equivalent alternatives and improvements within this scope are also within the scope of protection of the present invention.

Claims

1. A method for monitoring abnormal working conditions of an oil pump motor and its outlet pipeline based on sound recognition, characterized in that The following steps are involved: Step 1: Collect voiceprint information: The voiceprint sensor uses a patch-type bone conduction pickup, which is arranged on: the two sides of the motor housing, the two sides of the pump housing, the upper surface of the base, the oil pump outlet pipeline, the loading and unloading valve group, the oil tank drain valve, the pressure regulating valve body and the pressure regulating valve outlet pipeline; The temperature measuring resistors are arranged at: motor bearings and oil pump bearings; Pressure sensor layout: oil pump outlet, oil pressure tank and pressure regulating valve inlet; Liquid level sensor arrangement: oil sump and oil pressure tank; Step 1.

1. Arrange soundprint sensors on the pump body, motor, base, and valve outlet pipe of the oil pump motor; arrange temperature measuring resistors on the motor and oil pump bearings; arrange pressure sensors at the oil pump outlet, oil pressure tank, and pressure regulating valve inlet; arrange liquid level sensors on the oil sump and oil pressure tank; use the soundprint sensors to collect soundprint information, use the temperature measuring resistors to collect temperature information, use the pressure sensors to collect pressure information, and use the liquid level sensors to collect liquid level information; Step 1.2: The voiceprint information is processed by the front-end box and then transmitted to the edge computing gateway for preliminary screening. It is then transmitted to the server through the switch for data analysis. The temperature information, pressure information and liquid level information are sent to the local control unit LCU through the remote I / O module, and then transmitted to the server through the switch for comprehensive data analysis; Step 2: Extract voiceprint features from the voiceprint information and process the voiceprint features: Mel-frequency cepstral coefficients (MFCCs) are used to extract voiceprint information. The UBM method and the i-vector method are used to convert voiceprint features into model vectors that can be used for calculation and analysis. The probabilistic linear discriminant analysis (PLDA) method is used to compare and score the voiceprint features. Step 3: For voiceprint data with abnormal scores, a comprehensive logical judgment is performed based on multiple factors including pressure, liquid level, and temperature; Step 4: Output the final fault result; The steps to determine if the motor or oil pump is missing lubricating oil or the base is vibrating in step 3 are as follows: Through PLDA model analysis, a voiceprint feature fault sample library for the corresponding equipment is established. If the voiceprint feature data collected on-site successfully matches the fault sample library, the fault type is directly output; the voiceprint data that fails to match is scored by the PLDA model. If the score is abnormal, the fault is determined by combining relevant temperature data, and the fault sample is manually confirmed and marked to supplement the fault sample library. The steps are as follows: ①The motor lubricating oil is missing: Motor soundprint data S 1-1 or S 1-2 If the score is abnormal, analyze the rising trend of the motor bearing temperature T1 to =1min is the cycle, the current time t is the starting point, calculate the rising rate of T1 , the formula is as follows: ； If the rising rate of T1 in the last three cycles 、 、 The relationship between the three is: < , it can be judged that the motor bearing temperature rises abnormally, and the comprehensive voiceprint data score is abnormal, which can be determined to be the lack of motor lubricating oil; ②The oil pump is missing lubricating oil: Oil pump voiceprint data S 2-1 or S 2-2 If the score is abnormal, analyze the rising trend of the oil pump bearing temperature T2. If the rising rate of T2 in the last three cycles meets < , it can be determined that the oil pump bearing temperature rises abnormally, and the comprehensive soundprint data score is abnormal, which can be determined as the oil pump lubricating oil is missing; ③ Abnormal base vibration: If the soundprint data S3 score on the upper surface of the base is abnormal and the lubricating oil missing characteristics of the motor and oil pump do not match, it is determined that the base vibration is abnormal.

2. The method for monitoring abnormal working conditions of an oil pump motor and its outlet pipeline based on sound recognition according to claim 1, characterized in that: The MFCC extraction process in step 2 is as follows: 1) The collected voiceprint information is pre-emphasized through a high-pass filter and then divided into overlapping frames. A Hamming window function is applied to each frame to offset the assumption of infinite data made by the fast Fourier transform and reduce spectral leakage; 2) Perform a fast Fourier transform on each frame and calculate the power spectrum. The spectrum is extracted using a Mel filter bank. The spectrum is logarithmically calculated to obtain the energy. Then, a discrete cosine transform is applied to decorrelate the spectrum. 3) By extracting dynamic differential parameters, the MFCC feature vector consisting of N-dimensional MFCC parameters and frame energy is finally obtained; the dynamic differential parameters include first-order differential and second-order differential.

3. The method for monitoring abnormal working conditions of an oil pump motor and its outlet pipeline based on sound recognition according to claim 1, characterized in that: The process of calculating the UBM method in step 2 is as follows: First, a large amount of non-fault state voiceprint information is collected to train a universal background model (UBM). Then, a small amount of fault state voiceprint information is used to adjust the UBM parameters through an adaptive algorithm to obtain the target model parameters.

4. The method for monitoring abnormal working conditions of an oil pump motor and its outlet pipeline based on sound recognition according to claim 1, characterized in that: The steps for determining whether the pump blade is broken in step 3 are as follows: Through PLDA model analysis, a voiceprint feature fault sample library for the corresponding equipment is established. If the voiceprint feature data collected on-site successfully matches the fault sample library, the fault type is directly output; the voiceprint data that does not match is scored by the PLDA model. For those with abnormal scores, a fault determination is made based on multiple factors including temperature and pressure. The fault samples are manually confirmed and marked, and the fault sample library is supplemented. The steps are as follows: ④Pump blade fracture: Oil pump voiceprint data S 2-1 or S 2-2 The score is abnormal, the rising trend of the oil pump bearing temperature T2 is normal, and the oil pump outlet pressure P1 is lower than the normal loading and unloading pressure under loading and unloading conditions, that is, P 1加 P 加 And P 1卸 P 卸 , it can be determined that the oil pump efficiency is abnormal. Combined with the normal motor sound pattern data, it can be determined that the oil pump blade is broken.

5. The method for monitoring abnormal working conditions of an oil pump motor and its outlet pipeline based on sound recognition according to claim 4, characterized in that: The steps for determining whether the check valve function fails in step 3 are as follows: Through PLDA model analysis, a voiceprint feature fault sample library for the corresponding equipment is established. If the voiceprint feature data collected on-site successfully matches the fault sample library, the fault type is directly output; the voiceprint data that fails to match is scored by the PLDA model. If the score is abnormal, the fault is determined by combining relevant pressure data, and the fault sample is manually confirmed and marked to supplement the fault sample library. The steps are as follows: ⑤ Check valve function failure: When the oil pump motor stops working, the oil pump outlet pipe soundprint data S5 score is abnormal, and P1 0, and at the same time the soundprint data score of the oil pump motor is abnormal, it can be determined that the oil pump outlet check valve has failed, causing the oil pump motor to reverse.

6. The method for monitoring abnormal working conditions of an oil pump motor and its outlet pipeline based on sound recognition according to claim 5, characterized in that: In step 3, the steps to determine if the oil pump is frequently loaded and unloaded, the oil drain valve is leaking, or the servomotor pipeline is twitching are as follows: Through PLDA model analysis, a voiceprint feature fault sample library for the corresponding equipment is established. If the voiceprint feature data collected on-site successfully matches the fault sample library, the fault type is directly output; the voiceprint data that does not match is scored by the PLDA model. If the score is abnormal, a fault judgment is made based on multiple factors including pressure and liquid level. The fault sample is manually confirmed and marked, and the fault sample library is supplemented. The steps are as follows: ⑥ Frequent loading and unloading of the oil pump: The S6 score of the voiceprint data of the loading and unloading valve group is abnormal. Analyze the time interval between loading and unloading of the oil pump. , when the oil tank pressure is P3 When the pressure reaches 6.1 MPa, the oil pump starts loading and the time is recorded as , next time P3 When the pressure reaches 6.1 MPa, the time is recorded as ,but = - ,like ,in If it is the set value of loading and unloading interval, it is determined that the oil pump is loaded and unloaded frequently; ⑦ The oil drain valve is not closed tightly or leaks internally: The S4 score of the oil tank drain valve voiceprint data is abnormal, and the time it takes for the oil pump to load once Too long, that is > , is the setting value of the loading time, and the frequent loading and unloading characteristics have been matched. The total oil volume V of the hydraulic system is calculated by combining the liquid levels H1 and H2 of the pressure oil tank and the oil collection tank. If V=V0, where V0 is the initial total oil volume of the hydraulic system, it is determined that the oil drain valve is not closed tightly or there is internal leakage; ⑧Servomotor pipeline twitching: If the scores of the soundprint data S7 of the pressure-regulating valve body and the soundprint data S8 or S9 of the pressure-regulating valve outlet pipeline are abnormal, the frequent loading and unloading characteristics of the oil pump have been matched, and the total oil volume V=V0, it can be determined that the relay pipeline is twitching.

7. The method for monitoring abnormal working conditions of an oil pump motor and its outlet pipeline based on sound recognition according to claim 6, characterized in that: After the above faults are matched or confirmed, the host computer will output an alarm signal and fault message through human-computer interaction, and initiate corresponding preventive measures. After any of the faults ①-⑤ is confirmed to have occurred, the main and standby automatic switching function of the oil pump motor group should be activated, the standby oil pump motor group should be started, and the faulty group should be deactivated and marked as unavailable for maintenance. After fault ⑥ is confirmed to have occurred, only the alarm signal is output; If faults ⑥ and ⑦ occur simultaneously, it is considered that the oil drain valve is not closed tightly or is leaking internally, and an alarm signal is output, and the startup conditions are locked; Faults ⑥ and ⑧ occurred simultaneously, and were confirmed to be caused by twitching of the relay pipeline. An alarm signal was output, and the AGC was activated to adjust the unit output. According to the actual output of the unit, the output was increased or decreased in steps of 10MW to adjust the unit operating conditions. This process requires manual monitoring, and manual intervention and adjustment and on-site confirmation when necessary.

Images