CN109086888B - Automobile engine fault determination method and device based on voice recognition - Google Patents
Automobile engine fault determination method and device based on voice recognition Download PDFInfo
- Publication number
- CN109086888B CN109086888B CN201811033077.3A CN201811033077A CN109086888B CN 109086888 B CN109086888 B CN 109086888B CN 201811033077 A CN201811033077 A CN 201811033077A CN 109086888 B CN109086888 B CN 109086888B
- Authority
- CN
- China
- Prior art keywords
- fault
- automobile engine
- time
- frequency
- engine
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M15/00—Testing of engines
- G01M15/04—Testing internal-combustion engines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/14—Fourier, Walsh or analogous domain transformations, e.g. Laplace, Hilbert, Karhunen-Loeve, transforms
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
Abstract
The invention relates to a method and a device for judging automobile engine faults based on voice recognition, belongs to the technical field of engine fault detection, and solves the problems that in the prior art, the fault autonomous judgment is inaccurate, the running state information of an engine cannot be utilized in the judgment process, and the fault judgment needs to be carried out again in the later maintenance. The automobile engine fault judgment method disclosed by the invention adopts the hybrid neural network to identify faults according to AlexNet, and simultaneously uses LSTM to identify the running time state of the engine so as to assist in more accurate fault judgment, the number of layers of the AlexNet and the LSTM neural network is more than 9, and tests prove that the identification precision is higher and the judgment result is very accurate. In practical application, the method for judging the faults of the automobile engine can simply, quickly and reliably judge the common faults of the automobile engine, and remove the common faults in time, and fault judgment is not needed to be carried out again in later maintenance.
Description
Technical Field
The invention relates to the technical field of engine fault detection, in particular to a method and a device for judging automobile engine faults based on voice recognition.
Background
At present, in the daily use process of an automobile, due to the influence of natural and human factors, the problems of exhaust pipe faults, ignition system faults, throttle clamping stagnation, crankshaft main bearing faults and the like easily occur to an engine. Generally, a fault detection device is installed on an automobile engine, and whether the engine has a fault or not can be displayed through an instrument panel, but the prior art cannot indicate specific fault positions and fault types, so that a user cannot judge whether the automobile can continue to run or not, and meanwhile, the difficulty in maintaining the engine is increased, and the consumption of manpower and material resources is increased.
The failure of the automobile engine generally causes the sound of the automobile to change during operation, but at present, in the actual driving process, the sound of the engine is blocked due to the sealing and blocking of the automobile environment, or some failure sounds are fine so that a driver cannot hear the failure sounds in time. The existing autonomous engine sound identification method is usually to directly analyze faults from a time domain or a frequency domain, an adopted neural network architecture is generally a BP neural network (3-layer), the real-time running state information of an engine cannot be utilized in the judgment process, the identification precision is poor, and the fault judgment needs to be carried out again in the later maintenance.
Disclosure of Invention
In view of the foregoing analysis, embodiments of the present invention provide a method for determining a fault of an automobile engine based on voice recognition, so as to solve the problems in the prior art that an autonomous fault determination is not accurate, an engine operating state information cannot be utilized in a determination process, and a new fault determination is required for a later maintenance.
On one hand, the embodiment of the invention provides a method for judging the fault of an automobile engine based on voice recognition, which comprises the following steps:
collecting real-time monitoring sound data of an automobile engine;
performing time-frequency two-dimensional processing on the real-time monitoring sound data to obtain a time-frequency two-dimensional signal corresponding to the sound of the automobile engine;
inputting the time-frequency two-dimensional signal into a trained hybrid neural network, and judging whether an automobile engine has a fault and a specific fault position according to an output result of the hybrid neural network; the hybrid neural network comprises AlexNet and LSTM;
if yes, alarming and displaying the specific fault position;
if not, no fault is displayed.
The beneficial effects of the above technical scheme are as follows: the time-frequency two-dimensional signal is used as the input of the hybrid neural network, the information of two dimensions of time and frequency is reserved for audio scene data with poor structure, and compared with the traditional single-dimension sound identification method, the time-frequency two-dimensional signal has a better effect on audio information identification. Moreover, the technical scheme can be used for autonomously judging certain common faults of the engine, solves the problems that the autonomous judgment of the faults in the prior art is inaccurate, the running state information of the automobile engine cannot be utilized in the judgment process, and the fault judgment needs to be carried out again in the later maintenance, and meets the requirements of simply, quickly and reliably judging the faults of the automobile engine and timely eliminating the faults.
In another embodiment based on the above method, the step of training the hybrid neural network comprises:
acquiring a training set comprising N1 groups of automobile engine fault sound data and corresponding engine states and fault types; the automobile engine state comprises acceleration, deceleration and uniform speed, and the training set comprises all preset fault types of the automobile engine;
respectively carrying out time-frequency two-dimensional processing on the sound data of each group of automobile engines to obtain time-frequency two-dimensional signals corresponding to the sound of each group of automobile engines;
and inputting the time-frequency two-dimensional signals and the fault types corresponding to the sound of each group of automobile engines into AlexNet for training, and simultaneously inputting the time-frequency two-dimensional signals and the engine states corresponding to the sound of each group of automobile engines into LSTM for training to obtain a trained hybrid neural network.
The beneficial effects of the above technical scheme are: at present, a neural network architecture used by the existing engine fault judgment method based on voice recognition is a BP neural network, the number of layers is 3, AlexNet and LSTM neural networks adopted by the technical scheme have 9 or more layers, and the recognition precision is greatly improved.
Further, the time-frequency two-dimensional processing is carried out on the real-time monitoring sound data to obtain a time-frequency two-dimensional signal corresponding to the sound of the automobile engine, and the method comprises the following steps:
performing frame windowing on real-time monitoring sound data of an automobile engine to obtain N2Short-time sound data;
performing discrete Fourier transform on each short-time sound data to obtain corresponding short-time amplitude spectrum data;
and calculating a frequency spectrum density function of each short-time amplitude spectrum data, and taking the obtained frequency spectrum density function as a time-frequency two-dimensional signal corresponding to the sound of the automobile engine.
The beneficial effects of the further scheme are as follows: compared with a one-dimensional sound signal, the time-frequency two-dimensional signal not only retains the time characteristic, but also retains the frequency characteristic, contains more information, and is beneficial to the subsequent processes of mixed neural network feature extraction and identification.
Further, the step of inputting the time-frequency two-dimensional signal into the trained hybrid neural network and judging whether the automobile engine has a fault and the specific fault position according to the output result of the hybrid neural network comprises the following steps:
inputting the time-frequency two-dimensional signal into a trained hybrid neural network, obtaining the fault type and the corresponding occurrence probability I of the automobile engine through AlexNet, and obtaining the current state and the corresponding occurrence probability II of the automobile engine through LSTM;
if AlexNet has no output result or the occurrence probability of all fault types is 0, judging that the automobile engine works normally;
if the type of the fault in the AlexNet output result is only one, judging that the type of the fault occurs in the automobile engine;
if the AlexNet output result comprises at least two fault types, adding the first occurrence probability of each fault type and the corresponding second occurrence probability, and judging the fault type with the larger addition result of the automobile engine according to the addition result;
and determining the specific fault position of the automobile engine according to the fault type of the automobile engine.
The beneficial effects of the further scheme are as follows: when the sounds of different faults of the engine are close to each other, so that the faults are difficult to distinguish through AlexNet, the LSTM output result is used for auxiliary judgment to distinguish the faults, and therefore the result is more accurate.
Further, the obtaining includes N1The method comprises the following steps of grouping automobile engine fault sound data and corresponding training sets of engine states and fault types:
obtaining N1Grouping vehicle engine fault sound data; said N is1The group automobile engine fault sound data comprises all preset fault types;
calibrating the engine state and fault type corresponding to each group of automobile engine sound data to obtain N1Combining automobile engine sound data and corresponding training sets of engine states and fault types; the preset fault type comprises fault and deceleration in an acceleration stateFault in state, fault in uniform speed state.
The beneficial effects of the further scheme are as follows: and marking the sound data set of the automobile engine fault by using the existing experience to form training data, inputting AlexNet and LSTM neural networks for training, and obtaining a corresponding identification model of the engine real-time state and the engine fault.
Further, the faults in the acceleration state comprise blockage of an exhaust pipe, fault of an ignition system and weak high-voltage spark of a spark plug;
faults in the deceleration state comprise throttle valve clamping stagnation and crankshaft main bearing faults;
the faults in the uniform speed state comprise that the air cylinder is in short fire and the rotating shaft is unstable in uniform speed.
The beneficial effects of the further scheme are as follows: the method has the advantages that common faults of the automobile engine in different states are clarified, the fault type is associated with the working state of the automobile engine, and judgment is carried out by combining the fault type and the working state.
Further, a Hamming window is adopted as a window function of the frame windowing, and the frame overlapping is set to be 0.01-0.5 times of the window length.
The beneficial effects of the further scheme are as follows: because the actual automobile engine sound signal is long and the data volume is large, the very long sound data cannot be processed at one time without any need. Therefore, a segment of data is obtained by windowing and moving once for analysis, and the window length can also be used as the period size in the subsequent fourier transform. When the window is moved, the overlapping of the front frame and the rear frame is called frame overlapping, and the frame overlapping is set to be 0.01-0.5 times of the window length, so that the subsequent calculation is facilitated.
On the other hand, the embodiment of the invention provides an automobile engine fault judgment device based on voice recognition, which comprises an audio acquisition device, a voice recognition device and a central control display device which are sequentially connected;
the audio acquisition device is used for acquiring real-time monitoring sound data of the automobile engine, performing time-frequency two-dimensional processing on the real-time monitoring sound data, and inputting an acquired time-frequency two-dimensional signal corresponding to the sound of the automobile engine into the sound recognition device;
the sound recognition device is used for inputting the received time-frequency two-dimensional signal into a trained hybrid neural network and judging whether the automobile engine has a fault or not and the specific fault position according to the output result of the hybrid neural network; the hybrid neural network comprises AlexNet and LSTM;
and the central control display device is used for displaying the judgment result of the voice recognition device, alarming if a fault occurs, displaying the specific fault position, and displaying no fault if no fault occurs.
The beneficial effects of the above technical scheme are: the time-frequency two-dimensional signal is used as the input of the hybrid neural network, the information of two dimensions of time and frequency is reserved for audio scene data with poor structure, and compared with the traditional sound identification method, the time-frequency two-dimensional signal has a better effect on audio information identification. Moreover, the technical scheme can be used for autonomously judging certain specific faults of the engine, solves the problems that the fault autonomous judgment in the prior art is inaccurate, the running state information of the automobile engine cannot be utilized in the judgment process, and the fault judgment needs to be carried out again in the later maintenance, meets the requirement of simply, quickly and reliably judging the faults of the automobile engine, and is favorable for timely removal. Through the communication of the voice recognition device and the central control display device, the fault condition is reported to the user in time, and the driving safety of the automobile is also improved.
Further, the audio acquisition device comprises a sound receiving device and a signal processing module which are connected in sequence;
the sound receiving device is used for collecting real-time monitoring sound data of an automobile engine and sending the real-time monitoring sound data to the signal processing module;
and the signal processing module is used for performing time-frequency two-dimensional processing on the received real-time monitoring sound data and inputting the obtained time-frequency two-dimensional signal corresponding to the sound of the automobile engine into the sound recognition device.
The beneficial effects of the further scheme are as follows: the automobile engine sound is monitored in real time, the problem that faults cannot be found in time due to automobile sound insulation or weak engine fault sound can be solved, and the automobile driving safety is improved.
Further, the voice recognition device comprises AlexNet, LSTM and a judgment module;
the AlexNet is used for outputting the fault type of the automobile engine and the corresponding first occurrence probability according to the input time-frequency two-dimensional signal corresponding to the sound of the automobile engine;
the LSTM is used for outputting the current state of the automobile engine and the corresponding occurrence probability II according to the time-frequency two-dimensional signal corresponding to the input automobile engine sound;
and the judging module is used for judging whether the automobile engine judges whether a fault occurs or not and a specific fault position according to the output results of the AlexNet and the LSTM, if so, controlling the central control display device to give an alarm and display the specific fault position, and if not, controlling the central control display device to display no fault.
The beneficial effects of the further scheme are as follows: compared with the traditional voice recognition algorithm, the fault recognition is carried out by utilizing the AlexNet and LSTM neural networks, and the deep structure of the AlexNet and LSTM neural networks can effectively extract target characteristic information, so that the fault type of the automobile engine can be recognized more accurately.
In the invention, the technical schemes can be combined with each other to realize more preferable combination schemes. Additional features and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. The objectives and other advantages of the invention will be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings.
Drawings
The drawings are only for purposes of illustrating particular embodiments and are not to be construed as limiting the invention, wherein like reference numerals are used to designate like parts throughout.
FIG. 1 is a schematic diagram of the steps of a method for determining a fault of an automobile engine based on voice recognition according to embodiment 1 of the present invention;
FIG. 2 is a flow chart of a method for determining a fault of an automobile engine based on voice recognition according to embodiment 1 of the present invention;
FIG. 3 is a schematic diagram illustrating the steps of training a hybrid neural network according to embodiment 2 of the present invention;
FIG. 4 is a schematic diagram of a step of performing time-frequency two-dimensional processing on real-time monitored sound data according to embodiment 2 of the present invention;
FIG. 5 is a schematic diagram of a step of inputting a time-frequency two-dimensional signal into a trained hybrid neural network to determine whether a fault occurs and a specific fault location in accordance with embodiment 2 of the present invention;
FIG. 6 is a flowchart of a method for determining a fault of an automobile engine based on voice recognition according to embodiment 2 of the present invention;
FIG. 7 is a schematic connection diagram of an automobile engine fault determination device based on voice recognition according to embodiment 3 of the present invention;
fig. 8 is a schematic diagram of the components and connections of an audio acquisition device and a voice recognition device according to embodiment 3 of the present invention;
fig. 9 is a schematic diagram of the principle of an automobile engine fault determination device based on voice recognition according to embodiment 3 of the present invention.
Detailed Description
The accompanying drawings, which are incorporated in and constitute a part of this application, illustrate preferred embodiments of the invention and together with the description, serve to explain the principles of the invention and not to limit the scope of the invention.
Example 1
The invention discloses a method for judging automobile engine faults based on voice recognition, which comprises the following steps as shown in figure 1:
s1, collecting real-time monitoring sound data of an automobile engine through a sound collecting device.
And S2, performing time-frequency two-dimensional processing on the real-time monitoring sound data to obtain a time-frequency two-dimensional signal corresponding to the sound of the automobile engine.
S3, inputting the time-frequency two-dimensional signal into a trained hybrid neural network, and judging whether an automobile engine has a fault and a specific fault position according to the output result of the hybrid system; the hybrid neural network comprises AlexNet and LSTM. If yes, alarming and displaying the specific fault position; if not, no fault is displayed.
In this embodiment, the existing AlexNet includes 5 convolutional layers for feature extraction, 3 maximum pooling layers for implementing translational invariance of features, 2 fully-connected layers and a Softmax classifier for calculating probability and classifying engine sound types. The existing LSTM network structure is a common forgetting gate structure, and the output is the working state of an engine. The overall flow is shown in fig. 2.
Compared with the prior art, the automobile engine sound identification method provided by the embodiment adopts the time-frequency two-dimensional signal as the input of the hybrid neural network, so that the information of two dimensions of time and frequency can be reserved for audio scene data with poor structure. In addition, the method can be used for automatically judging common specific faults of the automobile engine, and solves the problems that the existing automobile engine fault automatic judgment is inaccurate, the running state information of the automobile engine cannot be utilized in the judgment process, and the fault judgment needs to be carried out again in the later maintenance.
Example 2
Based on the above embodiment, as shown in fig. 3, the step of training the hybrid neural network includes:
s01, obtaining N1Grouping automobile engine fault sound data and corresponding training sets of engine states and fault types; the engine state comprises acceleration, deceleration and uniform speed, and the training set comprises all fault types which can occur to the automobile engine.
And S02, respectively carrying out time-frequency two-dimensional processing on each group of automobile engine sound data to obtain time-frequency two-dimensional signals corresponding to each group of automobile engine sound.
And S03, inputting the time-frequency two-dimensional signals and the fault types corresponding to the sound of each group of automobile engines into AlexNet for training, and simultaneously inputting the time-frequency two-dimensional signals and the engine states corresponding to the sound of each group of automobile engines into LSTM for training to obtain a trained hybrid neural network.
Preferably, the iterative training of the above model is performed in a CAFFE architecture (open source) using a Back Propagation (BP) algorithm. The final result of the iterative training is that the loss function value converges to a stable value, at this time, it can be considered that the model training is finished, all parameters are fixed, and when new engine sound data is input, the model can classify the sound to judge which kind of fault the model belongs to.
Preferably, step S01 can be further refined into the following steps:
s011, acquiring N1And (4) assembling automobile engine fault sound data. Said N is1The group vehicle engine fault sound data includes all preset fault types. Specifically, audio data of a large number of automobile engines in various situations needs to be collected in advance (over ten thousand cases).
S012, calibrating the engine state and fault type corresponding to each group of automobile engine sound data to obtain the data including N1Combining automobile engine sound data and corresponding training sets of engine states and fault types; the preset fault types comprise faults in an acceleration state, faults in a deceleration state and faults in a constant speed state. Specifically, the practitioner may be invited to label to form the training data, e.g., label 0 indicates no failure, label 1 indicates a cylinder, label 2 indicates a piston failure, etc.
Preferably, the failure in the acceleration state includes exhaust pipe blockage, ignition system failure, weak spark plug high-voltage spark. The faults in the deceleration state comprise throttle valve clamping stagnation and crankshaft main bearing faults. The faults in the uniform speed state comprise that the air cylinder is in short fire and the rotating shaft is unstable in uniform speed. The position of the automobile engine fault can be determined according to the fault type, for example, the exhaust pipe blockage occurs at the position of the exhaust pipe, and details are not repeated below.
Preferably, as shown in fig. 4, the step S2 can be further refined into the following steps:
s21, performing frame windowing processing on real-time monitoring sound data of the automobile engineObtaining N2Short-time sound data. Specifically, a Hamming window is adopted as a window function for frame windowing, frame overlapping is set to be 0.01-0.5 times of the window length, and short-time sound data can be represented as xnAnd (m), n is a frame number, and m is a time number of the corresponding frame.
S22, performing Discrete Fourier Transform (DFT) on each short-time sound data to obtain corresponding short-time amplitude spectrum data
Wherein w is the frequency and N is the sequence length.
S23, calculating a frequency spectrum density function of each short-time amplitude spectrum data, and taking the obtained frequency spectrum density function P (n, w) as a time-frequency two-dimensional signal corresponding to the sound of the automobile engine
p(n,w)＝|X(n,w)|2＝(X(n,w)×(conj(X(n,w))) (2)
The horizontal axis of P (n, w) represents the time domain, the vertical axis represents the frequency domain, the value of a certain point represents the sound energy of the time frequency point, the horizontal axis represents the picture X direction and the vertical axis represents the picture Y direction in the image processing process similar to CNN, and the value of a certain point represents the gray scale value of the picture at the X, Y value.
Preferably, in step S3, the time-frequency two-dimensional signal is input into the trained hybrid neural network, and whether a fault occurs and a specific fault location are determined according to the output result of the hybrid neural network, as shown in fig. 5, the method may be further refined as follows:
and S31, inputting the time-frequency two-dimensional signal into a trained hybrid neural network, obtaining the fault type and the corresponding occurrence probability I of the automobile engine through AlexNet, and obtaining the current state and the corresponding occurrence probability II of the automobile engine through LSTM.
S32, if AlexNet has no output result or the occurrence probability of all fault types is 0, judging that the automobile engine works normally;
and S33, if only one fault type exists in the AlexNet output result, judging that the automobile engine has the fault of the type. At this time, the occurrence probability corresponding to the failure type is 100%.
And S34, if the AlexNet output result comprises at least two fault types, adding the first occurrence probability of each fault type and the corresponding second occurrence probability, and judging the fault type with the larger addition result of the automobile engine according to the addition result.
And S35, determining the specific fault position of the automobile engine according to the fault type with the larger summation result.
In implementation, different running states of an automobile engine, such as acceleration, deceleration, constant speed and the like, can be reflected acoustically, the flow chart of the automobile engine fault determination method is shown in fig. 6, AlexNet can only directly identify an input acoustic signal according to a hybrid neural network model, and cannot determine the relation between the signal and the running state of the automobile engine. The output of AlexNet is the probability value that the current input belongs to different categories, and under certain conditions, if the output result determines that the probabilities of a fault A (frequently occurring in an acceleration state) and a fault B (frequently occurring in a constant speed state) are similar, for example, about 40%, the fault identification can be assisted by using the running state information of the automobile engine, so that the LSTM is introduced to identify the running state of the automobile engine. LSTM introduces a storage element that when data has a time interval can relate this moment to a previous moment. If the LSTM network identifies that the automobile engine is in an acceleration state at the moment, the fault at the moment can be determined to belong to the category A and can be displayed and reported to a user or a driver in time.
Compared with the embodiment 1, the embodiment refines the structure of the hybrid neural network, refines the sound type of the common automobile engine, and further refines the fault identification by adopting the hybrid neural network. The technical scheme is established on the basis of a large number of test analysis, can effectively improve the automatic judgment capability and accuracy of the automobile engine, reduce the time and physical consumption of maintenance relocation diagnosis, and improve the driving safety of the automobile.
Example 3
An automobile engine fault determination device which performs determination by using the automobile engine fault determination method according to embodiment 2 includes, as shown in fig. 7, an audio acquisition device, a voice recognition device, and a central control display device which are connected in sequence.
And the audio acquisition device is used for acquiring real-time monitoring sound data of the automobile engine, performing time-frequency two-dimensional processing on the real-time monitoring sound data, and inputting the obtained time-frequency two-dimensional signals corresponding to the sound of the automobile engine into the sound recognition device.
The sound recognition device is used for inputting the received time-frequency two-dimensional signal into a trained hybrid neural network and judging whether the automobile engine has a fault or not and the specific fault position according to the output result of the hybrid neural network; the hybrid neural network comprises AlexNet and LSTM.
And the central control display device is used for displaying the judgment result of the voice recognition device, alarming if a fault occurs, displaying the specific fault position, and displaying no fault if no fault occurs.
Preferably, as shown in fig. 8, the audio acquisition device includes a sound receiving device and a signal processing module, which are connected in sequence. The sound receiving device is used for collecting real-time monitoring sound data of an automobile engine and sending the real-time monitoring sound data to the signal processing module; and the signal processing module is used for performing time-frequency two-dimensional processing on the received real-time monitoring sound data and inputting the obtained time-frequency two-dimensional signal corresponding to the automobile engine sound into the sound recognition device.
Preferably, as shown in fig. 8, the voice recognition device includes AlexNet, LSTM, and a determination module. And the AlexNet is used for outputting the fault type of the automobile engine and the corresponding first occurrence probability according to the input time-frequency two-dimensional signal corresponding to the sound of the automobile engine. And the LSTM is used for outputting the current state of the automobile engine and the corresponding occurrence probability II according to the time-frequency two-dimensional signal corresponding to the input automobile engine sound. And the judging module is used for judging whether the automobile engine judges whether a fault occurs or not and a specific fault position according to the output results of the AlexNet and the LSTM, if so, controlling the central control display device to give an alarm and display the specific fault position, and if not, controlling the central control display device to display no fault.
In the driving process of the automobile, the audio acquisition device converts acquired audio into a time-frequency two-dimensional signal matrix, and inputs the time-frequency two-dimensional signal matrix into the sound recognition device at a time interval of once a second. The voice recognition device is provided with a calculation board card storing a neural network recognition model, and the function can be realized by a sailing FPGA chip at present.
The vehicle central control display device is communicated with the voice recognition device. When the automobile engine is judged to be normal, the system displays no fault; when the automobile engine is judged to have a certain fault, the sound recognition device sends the specific fault type of the automobile engine to the automobile central control display device for displaying and alarming, and informs an automobile driver so as to eliminate potential safety hazards as soon as possible.
In the implementation, the schematic diagram of the engine failure determination device of the automobile is shown in fig. 9, for example, AlexNet detects an ignition system failure (43%), a cylinder misfire (41%), and determines whether to accelerate or to make a uniform speed according to LSTM, and if accelerating (71%), and making a uniform speed (12%), it can be determined as an ignition system failure. The criterion is that the probability values of the two networks are added to obtain a larger probability value.
Those skilled in the art will appreciate that all or part of the flow of the method implementing the above embodiments may be implemented by a computer program, which is stored in a computer readable storage medium, to instruct related hardware. The computer readable storage medium is a magnetic disk, an optical disk, a read-only memory or a random access memory.
The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope of the present invention are included in the scope of the present invention.
Claims (7)
1. A method for judging automobile engine faults based on voice recognition is characterized by comprising the following steps:
collecting real-time monitoring sound data of an automobile engine;
performing time-frequency two-dimensional processing on the real-time monitoring sound data to obtain a time-frequency two-dimensional signal corresponding to the sound of the automobile engine;
inputting the time-frequency two-dimensional signal into a trained hybrid neural network, and judging whether an automobile engine has a fault and a specific fault position according to an output result of the hybrid neural network; if yes, alarming and displaying the specific fault position; if not, displaying no fault;
the hybrid neural network comprises AlexNet and LSTM;
the step of training the hybrid neural network comprises:
obtaining includes N1Grouping automobile engine fault sound data and corresponding training sets of engine states and fault types; the automobile engine state comprises acceleration, deceleration and uniform speed, and the training set comprises all preset fault types of the automobile engine;
respectively carrying out time-frequency two-dimensional processing on the sound data of each group of automobile engines to obtain time-frequency two-dimensional signals corresponding to the sound of each group of automobile engines;
and inputting the time-frequency two-dimensional signals and the fault types corresponding to the sound of each group of automobile engines into AlexNet for training, and simultaneously inputting the time-frequency two-dimensional signals and the engine states corresponding to the sound of each group of automobile engines into LSTM for training to obtain a trained hybrid neural network.
2. The automobile engine fault determination method based on the voice recognition according to claim 1, wherein the time-frequency two-dimensional processing is performed on the real-time monitoring voice data to obtain a time-frequency two-dimensional signal corresponding to the voice of the automobile engine, and the method comprises the following steps:
performing frame windowing on real-time monitoring sound data of an automobile engine to obtain N2Short-time sound data;
performing discrete Fourier transform on each short-time sound data to obtain corresponding short-time amplitude spectrum data;
and calculating a frequency spectrum density function of each short-time amplitude spectrum data, and taking the obtained frequency spectrum density function as a time-frequency two-dimensional signal corresponding to the sound of the automobile engine.
3. The automobile engine fault determination method based on the voice recognition as claimed in claim 1, wherein the time-frequency two-dimensional signal is input into a trained hybrid neural network, and whether the automobile engine has a fault and a specific fault position is determined according to an output result of the hybrid neural network, comprising the following steps:
inputting the time-frequency two-dimensional signal into a trained hybrid neural network, obtaining the fault type and the corresponding occurrence probability I of the automobile engine through AlexNet, and obtaining the current state and the corresponding occurrence probability II of the automobile engine through LSTM;
if AlexNet has no output result or the occurrence probability of all fault types is 0, judging that the automobile engine works normally;
if the type of the fault in the AlexNet output result is only one, judging that the type of the fault occurs in the automobile engine;
if the AlexNet output result comprises at least two fault types, adding the first occurrence probability of each fault type and the corresponding second occurrence probability, and judging the fault type with the larger addition result of the automobile engine according to the addition result;
and determining the specific fault position of the automobile engine according to the fault type of the automobile engine.
4. The voice recognition-based automotive engine failure determination method according to claim 1, wherein the obtaining includes N1The method comprises the following steps of grouping automobile engine fault sound data and corresponding training sets of engine states and fault types:
obtaining N1Grouping vehicle engine fault sound data; said N is1The group automobile engine fault sound data comprises all preset fault types;
calibrating the engine state and fault type corresponding to each group of automobile engine sound data to obtain N1Combining automobile engine sound data and corresponding training sets of engine states and fault types; the preset fault types comprise faults in an acceleration state, faults in a deceleration state and faults in a constant speed state.
5. The voice recognition-based automotive engine failure determination method according to claim 4, characterized in that the failure in the acceleration state includes an exhaust pipe blockage, an ignition system failure, a spark plug high-voltage spark weakness;
faults in the deceleration state comprise throttle valve clamping stagnation and crankshaft main bearing faults;
the faults in the uniform speed state comprise that the air cylinder is in short fire and the rotating shaft is unstable in uniform speed.
6. The method for determining the engine fault of the automobile based on the voice recognition as claimed in claim 2, wherein the window function of the frame windowing processing adopts a Hamming window, and the frame stacking is set to be 0.01-0.5 times of the window length.
7. A fault judgment device of an automobile engine based on voice recognition is characterized by comprising an audio acquisition device, a voice recognition device and a central control display device which are sequentially connected;
the audio acquisition device is used for acquiring real-time monitoring sound data of the automobile engine, performing time-frequency two-dimensional processing on the real-time monitoring sound data, and inputting an acquired time-frequency two-dimensional signal corresponding to the sound of the automobile engine into the sound recognition device;
the sound recognition device is used for inputting the received time-frequency two-dimensional signal into a trained hybrid neural network and judging whether the automobile engine has a fault or not and the specific fault position according to the output result of the hybrid neural network; the hybrid neural network comprises AlexNet and LSTM;
the central control display device is used for displaying the judgment result of the voice recognition device, alarming if a fault occurs, displaying the specific fault position, and displaying no fault if no fault occurs;
the audio acquisition device comprises a sound receiving device and a signal processing module which are connected in sequence;
the sound receiving device is used for collecting real-time monitoring sound data of an automobile engine and sending the real-time monitoring sound data to the signal processing module;
the signal processing module is used for performing time-frequency two-dimensional processing on the received real-time monitoring sound data and inputting the obtained time-frequency two-dimensional signal corresponding to the sound of the automobile engine into the sound recognition device;
the voice recognition device comprises an AlexNet, an LSTM and a judgment module;
the AlexNet is used for outputting the fault type of the automobile engine and the corresponding first occurrence probability according to the input time-frequency two-dimensional signal corresponding to the sound of the automobile engine;
the LSTM is used for outputting the current state of the automobile engine and the corresponding occurrence probability II according to the time-frequency two-dimensional signal corresponding to the input automobile engine sound;
and the judging module is used for judging whether the automobile engine judges whether a fault occurs or not and a specific fault position according to the output results of the AlexNet and the LSTM, if so, controlling the central control display device to give an alarm and display the specific fault position, and if not, controlling the central control display device to display no fault.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811033077.3A CN109086888B (en) | 2018-09-05 | 2018-09-05 | Automobile engine fault determination method and device based on voice recognition |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811033077.3A CN109086888B (en) | 2018-09-05 | 2018-09-05 | Automobile engine fault determination method and device based on voice recognition |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109086888A CN109086888A (en) | 2018-12-25 |
| CN109086888B true CN109086888B (en) | 2020-10-27 |
Family
ID=64840641
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811033077.3A Active CN109086888B (en) | 2018-09-05 | 2018-09-05 | Automobile engine fault determination method and device based on voice recognition |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109086888B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11853863B2 (en) | 2019-08-12 | 2023-12-26 | Micron Technology, Inc. | Predictive maintenance of automotive tires |
Families Citing this family (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110441065B (en) * | 2019-07-04 | 2022-02-08 | 杭州华电江东热电有限公司 | Gas turbine on-line detection method and device based on LSTM |
| CN110534118B (en) * | 2019-07-29 | 2021-10-08 | 安徽继远软件有限公司 | Transformer/reactor fault diagnosis method based on voiceprint recognition and neural network |
| US20210049444A1 (en) * | 2019-08-12 | 2021-02-18 | Micron Technology, Inc. | Predictive maintenance of automotive engines |
| CN110749447B (en) * | 2019-11-27 | 2021-04-16 | 淮安信息职业技术学院 | Loader engine fault diagnosis method using big data |
| US11250648B2 (en) | 2019-12-18 | 2022-02-15 | Micron Technology, Inc. | Predictive maintenance of automotive transmission |
| CN111710329A (en) * | 2020-06-12 | 2020-09-25 | 重庆瑞尔科技发展有限公司 | Deep learning-based ship engine voiceprint fault discrimination method and system |
| CN112193250B (en) * | 2020-09-25 | 2022-03-18 | 三一专用汽车有限责任公司 | Method for monitoring vehicle running state, vehicle and computer readable storage medium |
| CN112326221B (en) * | 2020-10-28 | 2022-03-29 | 东风汽车集团有限公司 | Valve clamping stagnation simulation equipment and control method |
| CN112992179A (en) * | 2021-02-05 | 2021-06-18 | 安徽绿舟科技有限公司 | Recognition method for detecting faults of gas turbine based on voiceprint signals |
| CN113375933B (en) * | 2021-05-31 | 2022-09-09 | 中国矿业大学 | Fault diagnosis system and method for scraper conveyor |
| CN113566948A (en) * | 2021-07-09 | 2021-10-29 | 中煤科工集团沈阳研究院有限公司 | Fault audio recognition and diagnosis method for robot coal pulverizer |
| CN113571092A (en) * | 2021-07-14 | 2021-10-29 | 东软集团股份有限公司 | Method for identifying abnormal sound of engine and related equipment thereof |
| CN114112401A (en) * | 2021-11-10 | 2022-03-01 | 中国人民解放军陆军炮兵防空兵学院 | Engine fault diagnosis method of LSTM fault diagnosis model based on spectrogram |
| CN116380444B (en) * | 2023-06-05 | 2023-08-25 | 滨州鲁德曲轴有限责任公司 | Fault sound data processing system and processing method for crankshaft fault analysis |
| CN117332352A (en) * | 2023-10-12 | 2024-01-02 | 国网青海省电力公司海北供电公司 | Lightning arrester signal defect identification method based on BAM-AlexNet |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106404388A (en) * | 2016-09-13 | 2017-02-15 | 西安科技大学 | Scraper conveyor chain fluttering fault diagnosis method |
Family Cites Families (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| DE4421950C2 (en) * | 1993-12-09 | 1998-06-04 | Peter Mueller | Device for diagnosing and regulating a combustion or electric motor |
| US7660774B2 (en) * | 2005-05-31 | 2010-02-09 | Honeywell International Inc. | Nonlinear neural network fault detection system and method |
| CN104819846B (en) * | 2015-04-10 | 2017-03-22 | 北京航空航天大学 | Rolling bearing sound signal fault diagnosis method based on short-time Fourier transform and sparse laminated automatic encoder |
| CN107013449B (en) * | 2017-04-18 | 2018-08-24 | 山东万腾电子科技有限公司 | The method and system of voice signal identification compressor fault based on deep learning |
| CN106952649A (en) * | 2017-05-14 | 2017-07-14 | 北京工业大学 | Method for distinguishing speek person based on convolutional neural networks and spectrogram |
| CN108256556A (en) * | 2017-12-22 | 2018-07-06 | 上海电机学院 | Wind-driven generator group wheel box method for diagnosing faults based on depth belief network |
| CN108169639B (en) * | 2017-12-29 | 2020-05-19 | 南京康尼环网开关设备有限公司 | Method for identifying switch cabinet fault based on parallel long-time and short-time memory neural network |
-
2018
- 2018-09-05 CN CN201811033077.3A patent/CN109086888B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106404388A (en) * | 2016-09-13 | 2017-02-15 | 西安科技大学 | Scraper conveyor chain fluttering fault diagnosis method |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11853863B2 (en) | 2019-08-12 | 2023-12-26 | Micron Technology, Inc. | Predictive maintenance of automotive tires |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109086888A (en) | 2018-12-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109086888B (en) | Automobile engine fault determination method and device based on voice recognition | |
| WO2021232229A1 (en) | Virtual scene generation method and apparatus, computer device and storage medium | |
| KR102324776B1 (en) | Method for diagnosing noise cause of vehicle | |
| CN105679313A (en) | Audio recognition alarm system and method | |
| CN110108992B (en) | Cable partial discharge fault identification method and system based on improved random forest algorithm | |
| CN113870260A (en) | Welding defect real-time detection method and system based on high-frequency time sequence data | |
| US9311930B2 (en) | Audio based system and method for in-vehicle context classification | |
| CN112304614B (en) | Intelligent fault diagnosis method for end-to-end rolling bearing by adopting multi-attention mechanism | |
| CN109920448A (en) | A kind of identifying system and method for automatic driving vehicle traffic environment special type sound | |
| CN110176250B (en) | Robust acoustic scene recognition method based on local learning | |
| CN113566948A (en) | Fault audio recognition and diagnosis method for robot coal pulverizer | |
| CN110827793A (en) | Language identification method | |
| CN112735473A (en) | Method and system for identifying unmanned aerial vehicle based on voice | |
| CN111862951A (en) | Voice endpoint detection method and device, storage medium and electronic equipment | |
| CN111275108A (en) | Method for performing sample expansion on partial discharge data based on generation countermeasure network | |
| JP2012018066A (en) | Device for inspecting abnormality | |
| CN105931642A (en) | Speech recognition method, apparatus and system | |
| CN113763986A (en) | Air conditioner indoor unit abnormal sound detection method based on sound classification model | |
| CN113447902A (en) | Sea surveillance radar target identification method based on machine learning | |
| CN113962264A (en) | Aero-engine rotor system fault diagnosis algorithm based on deep learning | |
| CN113450827A (en) | Equipment abnormal condition voiceprint analysis algorithm based on compressed neural network | |
| CN111341343A (en) | Online updating system and method for abnormal sound detection | |
| CN115132197B (en) | Data processing method, device, electronic equipment, program product and medium | |
| CN113313982B (en) | Education system based on 5G network | |
| Čavor et al. | Vehicle Speed Estimation From Audio Signals Using 1D Convolutional Neural Networks |
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 |