CN113657221B - Power plant equipment state monitoring method based on intelligent sensing technology - Google Patents
Power plant equipment state monitoring method based on intelligent sensing technology Download PDFInfo
- Publication number
- CN113657221B CN113657221B CN202110889067.5A CN202110889067A CN113657221B CN 113657221 B CN113657221 B CN 113657221B CN 202110889067 A CN202110889067 A CN 202110889067A CN 113657221 B CN113657221 B CN 113657221B
- Authority
- CN
- China
- Prior art keywords
- fault
- sound
- signals
- model
- fault diagnosis
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 40
- 238000005516 engineering process Methods 0.000 title claims abstract description 25
- 238000012544 monitoring process Methods 0.000 title claims abstract description 23
- 230000005236 sound signal Effects 0.000 claims abstract description 69
- 238000003745 diagnosis Methods 0.000 claims abstract description 47
- 230000005856 abnormality Effects 0.000 claims abstract description 44
- 238000001514 detection method Methods 0.000 claims abstract description 42
- 230000002159 abnormal effect Effects 0.000 claims abstract description 36
- 230000008569 process Effects 0.000 claims abstract description 14
- 238000012549 training Methods 0.000 claims description 20
- 238000012706 support-vector machine Methods 0.000 claims description 13
- 239000013598 vector Substances 0.000 claims description 13
- 230000009467 reduction Effects 0.000 claims description 10
- 238000012545 processing Methods 0.000 claims description 9
- 238000001914 filtration Methods 0.000 claims description 6
- 238000001228 spectrum Methods 0.000 claims description 6
- 238000000605 extraction Methods 0.000 claims description 5
- 238000007781 pre-processing Methods 0.000 claims description 5
- 230000005484 gravity Effects 0.000 claims description 4
- 230000008447 perception Effects 0.000 claims description 4
- 238000009432 framing Methods 0.000 claims description 3
- 230000006870 function Effects 0.000 description 12
- 230000000694 effects Effects 0.000 description 5
- 230000008859 change Effects 0.000 description 4
- 238000007689 inspection Methods 0.000 description 4
- 238000012423 maintenance Methods 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 3
- 238000011161 development Methods 0.000 description 3
- 238000009826 distribution Methods 0.000 description 3
- 238000005457 optimization Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 230000003595 spectral effect Effects 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 238000007792 addition Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000015556 catabolic process Effects 0.000 description 1
- 238000013145 classification model Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 238000006731 degradation reaction Methods 0.000 description 1
- 230000030808 detection of mechanical stimulus involved in sensory perception of sound Effects 0.000 description 1
- 238000005265 energy consumption Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 230000037433 frameshift Effects 0.000 description 1
- 230000002045 lasting effect Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000013179 statistical model Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/48—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 specially adapted for particular use
- G10L25/51—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 specially adapted for particular use for comparison or discrimination
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computational Linguistics (AREA)
- Signal Processing (AREA)
- Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Human Computer Interaction (AREA)
- Acoustics & Sound (AREA)
- Multimedia (AREA)
- Testing And Monitoring For Control Systems (AREA)
- Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
Abstract
The invention provides a power plant equipment state monitoring method based on an intelligent sensing technology, which comprises the following steps of: s1, collecting sound signals in the running process of equipment; s2, carrying out abnormality detection on the sound signal by using an abnormality detection model, and inputting the abnormality signal into a fault diagnosis model when the abnormality signal is detected; s3, performing fault diagnosis on the abnormal signals by using a fault diagnosis model. According to the scheme, reliability judgment is carried out on the probability value of the fault type according to the fault diagnosis model, and the judgment result with low reliability is output to wait for manual calibration, so that the fault diagnosis model is continuously optimized in the use process, the fault type judgment accuracy can be continuously improved along with continuous use of the model, and the problem that fault diagnosis signals are missed to be judged in actual use due to insufficient fault type data can be solved.
Description
Technical Field
The invention belongs to the technical field of power plant equipment state monitoring, and particularly relates to a power plant equipment state monitoring method based on an intelligent sensing technology.
Background
Currently, equipment in a power plant often adopts a detection method, and the equipment is usually periodically checked and maintained, and mainly depends on manual experience and historical data. Because of the large discreteness of fault intervals of most equipment and spare parts, the method has certain disadvantages of limited maintenance effect and high cost. The inspection of a fixed period if too frequent results in not only a significant human cost, but also some unnecessary and even damaging maintenance activities. Secondly, if the period is not frequent enough, larger equipment failure risk and larger loss are brought. Therefore, by means of advanced technical means, real-time parameters of equipment operation are monitored and analyzed to judge whether the equipment has abnormality or fault, fault position and cause and fault degradation trend, so that reasonable maintenance opportunity is determined, accidents are eliminated in a sprouting state, maintenance cost is effectively reduced, and accident stopping rate is very necessary.
The power plant production environment has very large noise, the sound of different equipment is mixed in a relatively concentrated space, and as no good technical means is adopted for processing the sound signals in the past, the abnormal sound of the equipment can be heard only by the inspection staff with rich professional experience, and as the power plant equipment is developed towards the high, precise and sharp directions, the effective and accurate judgment on the running state of the equipment is difficult to be realized only by the inspection staff. In a practical system, the running states of the equipment under different working conditions are different, and when the working conditions are changed, if a fault occurs, the running states of the equipment are changed. Although the mechanism of sound signal generation of the device during the state change is relatively ambiguous, such sound signals tend to have non-stationary characteristics, so statistical model theory can be employed for analysis and processing. The change of the device state will often cause the change of the sound signal structure and different sound waveform diagrams, as shown in fig. 1, the running state of the device can be judged through the change of the sound signal characteristics of the device state, and even the device fault type and the occurrence position can be judged.
In recent years, with the development of technologies such as predictive control, kernel-bias least square, neural networks, support vector machines and the like and machine learning algorithms, and the development of the technologies is successfully applied in industry, a model of equipment fault diagnosis can be built by fully utilizing sufficient historical operation data of a power plant, and online analysis and intelligent early warning of equipment states are realized, however, with the high-speed development of power plant equipment, new fault types are gradually increased. In actual production, a large loss may be caused by missed judgment or misjudgment of a fault signal into a normal signal. The fault diagnosis model trained based on the historical data must distinguish the novel faults into a certain existing class. And normal operation data are far higher than fault type data in the operation of the equipment, and the model is easy to cause misjudgment or missed judgment on the fault data with very small quantity.
In order to accurately judge and early warn the fault type of power plant equipment, a new fault diagnosis method is required to be sought to solve the problem that fault signals are missed to be judged and misjudged by a fault diagnosis model in actual use due to insufficient fault type data.
Disclosure of Invention
The invention aims to solve the problems and provides a power plant equipment state monitoring method based on an intelligent sensing technology.
In order to achieve the above purpose, the present invention adopts the following technical scheme:
a power plant equipment state monitoring method based on intelligent perception technology comprises the following steps:
s1, collecting sound signals in the running process of equipment;
S2, carrying out abnormality detection on the sound signal by using an abnormality detection model, and inputting the abnormality signal into a fault diagnosis model when the abnormality signal is detected;
s3, performing fault diagnosis on the abnormal signals by using a fault diagnosis model.
In the power plant equipment state monitoring method based on the intelligent perception technology, the abnormality detection model is a single-class support vector machine, and is obtained by training in advance in the following manner:
A1. acquiring first sample data;
A2. Preprocessing a sound signal in the first sample data;
A3. Extracting the characteristics of the preprocessed sound signals to form characteristic vectors;
A4. performing dimension reduction treatment on the feature vector;
A5. training a single-class support vector machine by using the characteristics subjected to the dimension reduction treatment, and establishing an anomaly detection model;
in step S2, the collected sound signal is input into an abnormality detection model after data processing including preprocessing, feature extraction and dimension reduction processing is performed, so that the abnormality detection model performs abnormality detection on the sound signal.
In the power plant equipment state monitoring method based on the intelligent sensing technology, the sound signals in the first sample data are all normal signals;
Or the sound signal in the first sample data comprises a normal signal marked with a normal label and an abnormal signal marked with an abnormal label.
In the above power plant equipment state monitoring method based on the intelligent sensing technology, the step A2 specifically includes:
A21. Filtering the sound signal, and filtering low-frequency interference signals below 50Hz by using a high-pass filter;
A22. The sound is segmented and a sliding window framing process is used for each segment of sound.
In the above power plant equipment state monitoring method based on the intelligent sensing technology, in step A3, the extracted features include margin factors, pulse factors, skew factors in the time domain, and center of gravity frequency, mean square frequency and frequency features in the frequency spectrum.
In the above power plant equipment state monitoring method based on the intelligent sensing technology, in step A5, parameter optimization is performed by using an ant colony algorithm to obtain an optimal kernel function parameter and a balance parameter of a single-class support vector machine, and the optimal kernel function parameter and the balance parameter are substituted into the support vector machine to obtain a trained abnormality detection model, wherein the initial problem of the objective function is as follows:
Using a gaussian kernel function:
The conditions are satisfied: h (x i,xj)＝y(xi)Ty(xj) (3)
(1) The lagrangian pair problem is:
after each lambda i is solved, the discrimination function is obtained as follows:
(5) In the middle of
In the above formulas, l is the number of training samples;
Sigma is the required optimal kernel function parameter;
V is a trade-off parameter, which is a predefined percentage parameter estimation by the ant colony algorithm, and represents a compromise between the support vector and the wrong vector;
ρ is the compensation of the desired hyperplane in the feature space;
omega is the normal vector of the hyperplane required in the feature space;
ζ i is a relaxation variable;
Lambda i、λj is the lagrange multiplier and x i、xj is the sample in the original space.
In the above power plant equipment state monitoring method based on intelligent sensing technology, in step S3, the fault diagnosis model is used for outputting a fault type and a fault probability value determination of the corresponding fault type;
And step S3 further comprises:
S4, judging whether the fault probability value is higher than a preset probability value, if so, directly outputting the fault type, otherwise, executing the step S5;
S5, outputting the judged fault type and the fault probability value, and waiting for a worker to give a manual calibration result.
In the above-mentioned power plant equipment state monitoring method based on the intelligent sensing technology, in step S5, after the staff gives the artificial calibration result, the calibration result is used as a given tag of the corresponding sound signal, and the sound signal with the given tag is input to the abnormality detection model and/or the fault diagnosis model for training and updating.
In the above-mentioned power plant equipment state monitoring method based on intelligent sensing technology, in step S1, the sound signal in the running process of the equipment is collected at the first collection frequency;
in step S2, when an abnormal signal is detected, acquiring a sound signal in the running process of the equipment at a second acquisition frequency, and simultaneously inputting the abnormal signal and the sound signal acquired at the second acquisition frequency into a fault diagnosis model;
the second acquisition frequency is greater than the first acquisition frequency.
In the above-mentioned power plant equipment state monitoring method based on the intelligent sensing technology, in step S2, the abnormality detection model simultaneously detects the sound signal during the operation of the equipment collected at the second collection frequency, and when the sound signals which last for the preset times/time are detected as normal sound signals, the sound signals are restored to the first collection frequency.
The invention has the advantages that:
1. the problem of failure diagnosis signal missing judgment in actual use caused by insufficient failure type data can be solved;
2. and according to the fault diagnosis model, reliability judgment is carried out on the probability value of the fault type, and a judgment result with low reliability is output to wait for manual calibration, so that the fault diagnosis model is continuously optimized in the use process, and the fault type judgment accuracy can be continuously improved along with continuous use of the model.
Drawings
FIG. 1 is a sound waveform diagram of a device in various operating conditions;
FIG. 2 is a training flow chart of an anomaly detection model in the present invention;
FIG. 3 is a training flow chart of the fault diagnosis model of the present invention;
FIG. 4 is a flow chart of a judging method of the power plant equipment state monitoring method based on the intelligent sensing technology;
FIG. 5 is a graph of the performance of the ROC curve evaluation anomaly detection model-AUC (Area Under Curve).
Detailed Description
The invention is described in further detail below with reference to the drawings and the detailed description.
As shown in fig. 2, the present embodiment discloses a power plant equipment status monitoring method based on an intelligent sensing technology, and firstly prepares a trained abnormality detection model and a fault diagnosis model.
The anomaly detection model adopts a single-class support vector machine, and is mainly obtained through the following training:
A1. acquiring first sample data; the sound signals in the first sample data may be normal signals; a normal signal labeled normal and an abnormal signal labeled abnormal may also be included.
A2. Preprocessing a sound signal in first sample data, firstly filtering the sound signal, and filtering a low-frequency interference signal below 50Hz by using a high-pass filter; the sound is divided into 6s each section, and each section of sound is subjected to frame division treatment by using a sliding window, wherein the window length is 0.25s, and the window is shifted by 50%, namely, half of the window length is shifted each time.
A3. Extracting the characteristics of the preprocessed sound signals to form characteristic vectors: and extracting the statistical characteristics of all frame signals in the time domain and the characteristics of the frequency spectrum distribution, such as margin factors, pulse factors, deflection factors in the time domain, center of gravity frequency, mean square frequency and frequency characteristics in the frequency spectrum, wherein fs is set as a sampling frequency, continuous sound signals s (t) per second are sampled and discretized into s (N), the frame length of framing processing is set as N, the frame shift is set as N/2, and the total frame number is set as N F. The statistical characteristics of the ith frame signal s i (n) in the time domain, the margin factor, and the extraction formula is:
The impulse factor is extracted by the following formula:
the skew factor is extracted by the following formula:
s (N) (n=1, 2,) N is the time domain signal and N is the signal sample length.
The signal power spectrum reflects the random distribution of signal energy, i.e. the analysis of the characteristics of the signal from the frequency components in the signal and the energy magnitudes of the frequency components. Performing Fourier transform on the signal s (n) to obtain the distribution information of the signal s (n) on the frequency spectrum, wherein the formula of the fast Fourier transform is as follows:
Where s i (k) is STFT (short time fourier transform) of the i-th frame signal s i (n).
The characteristics of the i-th frame signal s i (k) in the frequency domain,
The center of gravity frequency is extracted as follows:
The mean square frequency is extracted by the following formula:
The frequency characteristics are extracted by the following formula:
y (K) (k=1, 2,., K) is the spectral value of the signal s (n), K is the number of spectral lines, and f k is the frequency value. The above characteristic parameters of the signal in the time domain and the frequency domain are formed into a 6-dimensional vector which is used as the characteristic vector of a section of sound signal.
A4. and performing dimension reduction processing on the feature vector: finding out main features, and replacing original data with the main features;
A5. the feature training single-class support vector machine subjected to dimension reduction is utilized to establish an anomaly detection model, and the anomaly detection model specifically comprises the following steps:
carrying out normalization processing on the feature parameters after dimension reduction; performing parameter optimization by using an ant colony algorithm to obtain optimal kernel function parameters and balance parameters of a single-class support vector machine, and constructing and solving the optimal problem; the initial problem of the objective function is:
Using a gaussian kernel function:
the conditions are satisfied: h (x i,xj)＝y(xi)Ty(xj) (10)
(8) The lagrangian pair problem is:
after each lambda i is solved, the discrimination function is obtained as follows:
(12) In the middle of
In the above formulas, l is the number of training samples;
Sigma is the required optimal kernel function parameter;
v.epsilon.0, 1 is a trade-off parameter, which is a predefined percentage parameter estimate, representing a trade-off between support vector and misvector;
ρ is the compensation of the desired hyperplane in the feature space;
omega is the normal vector of the hyperplane required in the feature space;
ζ i is a relaxation variable;
Lambda i、λj is the lagrange multiplier and x i、xj is the sample in the original space.
Substituting the optimal values of the model parameters (v, sigma) of the single-class support vector machine obtained by ant colony optimization into the support vector machine to obtain a trained anomaly detection model.
The fault diagnosis model adopts a one-dimensional convolutional neural network consisting of an input layer, a feature extraction layer and a classification layer. As shown in fig. 3, the abnormal signal of the marked fault type is input into the fault diagnosis model for training, and the trained fault diagnosis model can be obtained. The abnormal signal can be prepared in advance by a worker, or the abnormal detection model can be used for detecting the second sample data, namely, the second sample data is input into the judging function (12) after being processed, so that a training sample set D of the abnormal signal is obtained, then the worker classifies the data in the training sample set according to different fault types, the data are respectively marked as D 1,D2,...,Dk, and classification marks are made.
The first sample data and the second sample data may use different sample data or may use the same sample data.
When the device is put into use, as shown in fig. 4, the collected device original signal is input to an abnormality detection model to perform abnormality judgment, and then the abnormality signal is input to a fault diagnosis model to perform fault classification by the fault diagnosis model. The specific method comprises the following steps:
S1, acquiring sound signals in the running process of equipment at a first acquisition frequency, wherein the first acquisition frequency can be 10 minutes once, and each acquisition time is 6 seconds; the sound sensor on the equipment to be tested can be used for collecting the sound signals in the running process of the equipment, the magnetic mounting mode can directly acquire the sound inside the equipment, the influence of environmental noise on the signals is reduced, the intelligent sound sensor simulating human ear hearing is preferably utilized, and the problem of delay in collection and transmission of the sound signals is solved by applying the 5G technology of the Internet of things.
S2, inputting the collected sound signals into an abnormality detection model after pretreatment, feature extraction and dimension reduction treatment so as to detect abnormality of the sound signals by the abnormality detection model, and inputting the abnormality signals into a fault diagnosis model when the abnormality signals are detected;
s3, performing fault diagnosis on the abnormal signals by using a fault diagnosis model.
Since the abnormality detection model cannot distinguish one hundred percent of normal signals from abnormal signals, it is possible that some normal signals in which the apparatus operates in an extreme condition are misjudged as abnormal signals (the abnormality diagnosis model does not judge the abnormal signals as normal signals, but it is possible that the normal signals are judged as abnormal signals), and thus the detected data may contain a known type of failure and an abnormal type (unknown type of failure or normal), the present embodiment adds judgment conditions after the failure recognition model for distinguishing the abnormal type.
Specifically, in step S3, the fault diagnosis model outputs not only the fault type but also the fault probability value of the corresponding fault type;
And step S3 further comprises:
s4, judging whether the fault probability value is higher than a preset probability value, if so, 0.9, wherein the preset probability value is used as a critical point capable of accurately judging, if so, judging the fault type accurately, and directly outputting the fault type, otherwise, executing the step S5;
S5, considering that the fault type cannot be accurately judged, outputting the judged fault type and fault probability value, and waiting for personnel to give a manual calibration result.
When the staff gives the artificial calibration result, the calibration result is used as a given label of the corresponding sound signal, and the sound signal with the given label is input into the abnormality detection model and/or the fault diagnosis model for training and updating. If the calibration is normal, the relevant sound signals are placed in a normal sample database, the abnormal detection model is trained and updated, and if the calibration is abnormal, the sound signals are placed in a fault sample database, and the fault diagnosis model is trained and updated.
Preferably, when an abnormal signal is detected, the sound signal during the operation of the apparatus is collected at the second collection frequency, and the abnormal signal and the sound signal collected at the second collection frequency are simultaneously input to the fault diagnosis model; the second acquisition frequency may be 1 minute once for a period of 6 seconds. Under normal conditions, the data are collected for 10-30 minutes once, the collection frequency is increased when the primary judgment is abnormal, the energy consumption can be saved, the requirement of frequent collection is avoided, the continuous collection of the data in abnormal time can be met, and the data collection device has similar inspection effects.
Further, the abnormality detection model simultaneously detects the sound signals in the operation process of the collection device at the second collection frequency, and when the sound signals lasting for the preset times/time are detected as normal sound signals, the sound signals are restored to the first collection frequency. When the sound signal is abnormal, the sound signal is continuously collected, so that the fault state is diagnosed by the model, and when the sound signal is normal, the sound signal is not required to be collected frequently and is recovered to be collected at long intervals.
And evaluating the OCSVM abnormality detection model by adopting ROC-AUC model evaluation indexes. First, the test data is manually labeled, and normal sound and abnormal sound data are manually identified. Next, the test data of the tag is input into the ROC-AUC evaluation program, and a result chart of the ROC is obtained as shown in fig. 5. In the ROC result graph, the closer the curve is to the upper right corner of the coordinate system, the higher the accuracy of the model is, and the better the effect is. Considering that the ROC curve itself cannot intuitively describe the performance of a classifier, and the AUC represents the area under the ROC curve, the ROC curve is mainly used for measuring the generalization performance of the model, namely the classification effect, the value is used as a number value of [0,1], the closer the value is to 1, the evaluated model has comparability, and quantitative comparison can be performed. The AUC value of the model is 0.891, which proves that the model can finish the task excellently in the detection of sound abnormality in specific situations.
According to the scheme, firstly, abnormality detection is carried out and then fault diagnosis is carried out, only abnormal data are used for training a fault diagnosis model, and the influence of an unbalanced sample on the identification accuracy of the classification model is avoided; the abnormality detection model can detect all abnormal signals, so that the problem that the fault diagnosis model can not judge faults in actual use due to insufficient fault type data can be well solved; and judging whether the model is reliable or not according to the probability value output by diagnosis, waiting for manual calibration for unreliable results and further updating the training model, and continuously improving the model diagnosis accuracy along with continuous use of the model.
The specific embodiments described herein are offered by way of example only to illustrate the spirit of the invention. Those skilled in the art may make various modifications or additions to the described embodiments or substitutions thereof without departing from the spirit of the invention or exceeding the scope of the invention as defined in the accompanying claims.
Although terms such as sound signal, abnormality detection amount model, fault diagnosis model, single-class support vector machine, normal signal, abnormal signal, etc. are used more herein, the possibility of using other terms is not excluded. These terms are used merely for convenience in describing and explaining the nature of the invention; they are to be interpreted as any additional limitation that is not inconsistent with the spirit of the present invention.
Claims (5)
1. The power plant equipment state monitoring method based on the intelligent sensing technology is characterized by comprising the following steps of:
s1, acquiring sound signals in the running process of equipment at a first acquisition frequency;
S2, carrying out abnormality detection on the sound signals by using an abnormality detection model, and when the abnormality signals are detected, acquiring the sound signals in the running process of the equipment at a second acquisition frequency, and simultaneously inputting the abnormality signals and the sound signals acquired at the second acquisition frequency into a fault diagnosis model;
The second acquisition frequency is greater than the first acquisition frequency;
The abnormal detection model detects the sound signals in the running process of the second acquisition frequency acquisition equipment at the same time, and when the sound signals which last for preset times/time are detected as normal sound signals, the sound signals are restored to the first acquisition frequency;
s3, performing fault diagnosis on the abnormal signal by using a fault diagnosis model, and outputting a fault type and a fault probability value judgment of the corresponding fault type by using the fault diagnosis model;
The abnormality detection model and the fault diagnosis model are both trained models, and the fault diagnosis model is obtained by training abnormality data;
s4, judging whether the fault probability value is higher than a preset probability value, wherein the preset probability value is used as a critical point capable of accurately judging whether the fault probability value is higher than the preset probability value, if so, judging the fault type accurately, and outputting the fault type directly, otherwise, executing the step S5;
s5, outputting the judged fault type and fault probability value, and waiting for a worker to give a manual calibration result;
And after the personnel give the artificial calibration result, taking the calibration result as a given label of the corresponding sound signal, if the calibration is normal, putting the corresponding sound signal into a normal sample database, training and updating the abnormal detection model, and if the calibration is a certain fault, putting the corresponding sound signal into a fault sample database, and training and updating the fault diagnosis model.
2. The method for monitoring the state of power plant equipment based on the intelligent perception technology according to claim 1, wherein the abnormality detection model is a single-class support vector machine, and is obtained by training in advance in the following manner:
A1. acquiring first sample data;
A2. Preprocessing a sound signal in the first sample data;
A3. Extracting the characteristics of the preprocessed sound signals to form characteristic vectors;
A4. performing dimension reduction treatment on the feature vector;
A5. training a single-class support vector machine by using the characteristics subjected to the dimension reduction treatment, and establishing an anomaly detection model;
in step S2, the collected sound signal is input into an abnormality detection model after data processing including preprocessing, feature extraction and dimension reduction processing is performed, so that the abnormality detection model performs abnormality detection on the sound signal.
3. The intelligent perception technology-based power plant equipment state monitoring method according to claim 2, wherein the sound signals in the first sample data are all normal signals;
Or the sound signal in the first sample data comprises a normal signal marked with a normal label and an abnormal signal marked with an abnormal label.
4. The method for monitoring the state of power plant equipment based on the intelligent sensing technology according to claim 2, wherein the step A2 is specifically:
A21. Filtering the sound signal, and filtering low-frequency interference signals below 50Hz by using a high-pass filter;
A22. The sound is segmented and a sliding window framing process is used for each segment of sound.
5. The method for monitoring the status of power plant equipment based on the intelligent sensing technology according to claim 4, wherein in the step A3, the extracted features include margin factors, pulse factors, skew factors in time domain and center of gravity frequencies, mean square frequencies and frequency features in frequency spectrum.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110889067.5A CN113657221B (en) | 2021-08-04 | 2021-08-04 | Power plant equipment state monitoring method based on intelligent sensing technology |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110889067.5A CN113657221B (en) | 2021-08-04 | 2021-08-04 | Power plant equipment state monitoring method based on intelligent sensing technology |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113657221A CN113657221A (en) | 2021-11-16 |
| CN113657221B true CN113657221B (en) | 2024-08-23 |
Family
ID=78490281
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110889067.5A Active CN113657221B (en) | 2021-08-04 | 2021-08-04 | Power plant equipment state monitoring method based on intelligent sensing technology |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113657221B (en) |
Families Citing this family (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114692674B (en) * | 2022-02-21 | 2024-09-17 | 武汉船用电力推进装置研究所（中国船舶重工集团公司第七一二研究所） | Ship propulsion system fault judging method and system |
| CN115078912A (en) * | 2022-06-22 | 2022-09-20 | 中车青岛四方机车车辆股份有限公司 | Method and system for detecting abnormity of roof high-voltage cable connector in real time and train |
| CN115410595A (en) * | 2022-08-17 | 2022-11-29 | 广州瀚信通信科技股份有限公司 | Machine room operation and maintenance method and system based on 5G NB-IoT |
| CN115468707A (en) * | 2022-08-31 | 2022-12-13 | 南方电网调峰调频发电有限公司 | Method for monitoring gas leakage of power plant sealing equipment by combining listening and vibration |
| CN115588265B (en) * | 2022-12-12 | 2023-04-07 | 华能酒泉风电有限责任公司 | Intelligent monitoring system of wind power plant |
| CN116384980B (en) * | 2023-05-25 | 2023-08-25 | 杭州青橄榄网络技术有限公司 | Repair reporting method and system |
| CN116720073A (en) * | 2023-08-10 | 2023-09-08 | 江苏金恒信息科技股份有限公司 | Abnormality detection extraction method and system based on classifier |
| CN117409816B (en) * | 2023-12-14 | 2024-03-26 | 湖南华夏特变股份有限公司 | Equipment fault detection method and system based on sound signals |
| CN119181384B (en) * | 2024-11-25 | 2025-01-24 | 中国空气动力研究与发展中心高速空气动力研究所 | Method for judging equipment abnormality based on forward sample and on-line extraction of audio features |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112233695A (en) * | 2020-10-13 | 2021-01-15 | 张鹏 | Oiling machine abnormal sound analysis and fault early warning system based on artificial intelligence and big data |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102122133B (en) * | 2011-01-21 | 2012-10-10 | 北京工业大学 | Self-adaption wavelet neural network abnormity detection and fault diagnosis classification system and method |
| CN107505133B (en) * | 2017-08-10 | 2019-05-28 | 滁州学院 | The probability intelligent diagnosing method of rolling bearing fault based on adaptive M RVM |
| JP7149197B2 (en) * | 2019-02-06 | 2022-10-06 | 株式会社日立製作所 | ABNORMAL SOUND DETECTION DEVICE AND ABNORMAL SOUND DETECTION METHOD |
| CN112087703A (en) * | 2020-08-14 | 2020-12-15 | 陕西千山航空电子有限责任公司 | Audio acquisition fault diagnosis method for audio monitor |
-
2021
- 2021-08-04 CN CN202110889067.5A patent/CN113657221B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112233695A (en) * | 2020-10-13 | 2021-01-15 | 张鹏 | Oiling machine abnormal sound analysis and fault early warning system based on artificial intelligence and big data |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113657221A (en) | 2021-11-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113657221B (en) | Power plant equipment state monitoring method based on intelligent sensing technology | |
| CN113305645B (en) | Numerical control machine tool cutter residual life prediction method based on hybrid neural model | |
| Yan et al. | Fisher’s discriminant ratio based health indicator for locating informative frequency bands for machine performance degradation assessment | |
| CN113076975A (en) | Dam safety monitoring data anomaly detection method based on unsupervised learning | |
| CN111665066B (en) | Equipment fault self-adaptive upper and lower early warning boundary generation method based on convolutional neural network | |
| CN118209830A (en) | Intelligent monitoring method and system for cable partial discharge abnormality | |
| CN115409131B (en) | Production line abnormity detection method based on SPC process control system | |
| CN112173636B (en) | Method for detecting faults of belt conveyor carrier roller by inspection robot | |
| CN110823576A (en) | Generative Adversarial Network-Based Mechanical Anomaly Detection Method | |
| CN111678699B (en) | A method and system for early fault monitoring and diagnosis of rolling bearing | |
| CN107844067B (en) | A kind of gate of hydropower station on-line condition monitoring control method and monitoring system | |
| CN108956111B (en) | Abnormal state detection method and detection system for mechanical part | |
| US20210116504A1 (en) | Method for monitoring circuit breaker and apparatus and internet of things using the same | |
| CN118050591A (en) | Distribution network 10KV overhead line fault early warning and positioning system based on AI | |
| CN118311934B (en) | Electric automatization control system | |
| CN112669879A (en) | Air conditioner indoor unit noise anomaly detection method based on time-frequency domain deep learning algorithm | |
| CN117807538A (en) | Fault detection method based on dynamic threshold and trend prediction | |
| CN109141625B (en) | Online state monitoring method for ball screw pair | |
| CN117435908A (en) | Multi-fault feature extraction method for rotary machine | |
| CN118130984A (en) | Cable partial discharge fault real-time monitoring method based on data driving | |
| US20210149387A1 (en) | Facility failure prediction system and method for using acoustic signal of ultrasonic band | |
| CN113177646A (en) | Power distribution equipment online monitoring method and system based on self-adaptive edge proxy | |
| CN109598309B (en) | Detection system and monitoring method of metal packaging punching machine | |
| CN118536048B (en) | Switchgear insulation status monitoring and management system | |
| CN207992717U (en) | A kind of gate of hydropower station on-line condition monitoring system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |