CN117009770A - Bearing fault diagnosis method based on SDP and visual transducer codes - Google Patents
Bearing fault diagnosis method based on SDP and visual transducer codes Download PDFInfo
- Publication number
- CN117009770A CN117009770A CN202311106774.8A CN202311106774A CN117009770A CN 117009770 A CN117009770 A CN 117009770A CN 202311106774 A CN202311106774 A CN 202311106774A CN 117009770 A CN117009770 A CN 117009770A
- Authority
- CN
- China
- Prior art keywords
- sdp
- image block
- layer
- attention
- bearing
- 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.)
- Pending
Links
- 238000003745 diagnosis Methods 0.000 title claims abstract description 79
- 230000000007 visual effect Effects 0.000 title claims abstract description 51
- 238000000034 method Methods 0.000 title claims abstract description 39
- 238000010586 diagram Methods 0.000 claims abstract description 35
- 239000013598 vector Substances 0.000 claims abstract description 13
- 230000006870 function Effects 0.000 claims description 33
- 239000011159 matrix material Substances 0.000 claims description 18
- 230000004913 activation Effects 0.000 claims description 17
- 238000013507 mapping Methods 0.000 claims description 14
- 230000007246 mechanism Effects 0.000 claims description 13
- 238000000605 extraction Methods 0.000 claims description 11
- 238000012545 processing Methods 0.000 claims description 9
- 238000010606 normalization Methods 0.000 claims description 7
- 230000003321 amplification Effects 0.000 claims description 6
- 238000003199 nucleic acid amplification method Methods 0.000 claims description 6
- 238000007781 pre-processing Methods 0.000 claims description 5
- 230000007774 longterm Effects 0.000 abstract description 8
- 230000008569 process Effects 0.000 description 11
- 230000009466 transformation Effects 0.000 description 11
- 238000013528 artificial neural network Methods 0.000 description 6
- 230000008901 benefit Effects 0.000 description 5
- 238000013135 deep learning Methods 0.000 description 5
- 238000012549 training Methods 0.000 description 5
- 238000005096 rolling process Methods 0.000 description 3
- 238000010183 spectrum analysis Methods 0.000 description 3
- 230000001629 suppression Effects 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 125000004122 cyclic group Chemical group 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000035945 sensitivity Effects 0.000 description 2
- 241000533950 Leucojum Species 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000006698 induction Effects 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 238000012360 testing method 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
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M13/00—Testing of machine parts
- G01M13/04—Bearings
- G01M13/045—Acoustic or vibration analysis
-
- 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
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- 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
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computational Linguistics (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Software Systems (AREA)
- Acoustics & Sound (AREA)
- Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
Abstract
The invention particularly relates to a bearing fault diagnosis method based on SDP and visual transducer codes, which comprises the following steps: converting the vibration signal into an SDP diagram; inputting the SDP graph into a trained fault diagnosis model to output a corresponding fault type predicted value: splitting the SDP graph into a series of SDP image block sequences; linearly adding position embedding in the vector of the SDP image block sequence to obtain a position embedded image block sequence; capturing local information and global information in an SDP image block sequence and learning dependency relations among different parts in the SDP image block sequence to obtain an SDP image block sequence coding representation; probability values of various fault types are obtained through the MLP head layer, and then the fault type corresponding to the maximum probability value is used as a fault type prediction value. According to the invention, the SDP algorithm and the visual transducer code are effectively fused, so that the characteristic information of bearing faults can be effectively extracted from complex vibration signals, and meanwhile, the long-term dependency relationship in the time sequence of the bearing vibration signals can be better captured.
Description
Technical Field
The invention relates to the field of bearing fault diagnosis and deep learning, in particular to a bearing fault diagnosis method based on SDP and visual transducer codes.
Background
Rolling bearings are one of the most widely used components in modern mechanical devices, and rely on rolling contact between the primary elements to support the rotating parts, which support the shaft while allowing the shaft to be flexibly rotated while being supported. Through practical researches, faults of mechanical equipment such as rotary machinery, induction motors, gearboxes and the like are mostly caused by rolling bearings, and after the bearing state is monitored and diagnosed, the accident rate and the maintenance cost are greatly reduced. When the bearing is monitored by the sensor and judged by the model, the manpower can be greatly reduced, and more accurate information feedback is obtained, so that the advanced bearing fault diagnosis system can greatly improve the benefit of actual engineering.
In recent years, due to the continuous development of deep learning technology, many researchers consider combining the bearing failure field with a deep learning algorithm. The Chinese patent publication No. CN111198098A discloses a wind driven generator bearing fault prediction method based on an artificial neural network, which comprises the following steps: collecting historical data of the running of the wind driven generator bearing, and preprocessing the data; performing frequency transformation by using improved stable wavelet packet transformation, thereby performing frequency bandwidth separation and extracting fault characteristic frequency values; training an Elman artificial neural network by using a training set to obtain a neural network model; and performing fault prediction on the input pair real-time data. According to the existing scheme, the prediction of bearing faults is realized through an artificial neural network.
The applicant has found that the vibration signal of the bearing is one of the signals that best reflects its fault condition. For large-scale mechanical equipment, the running condition and the environment of the bearing are complex, and then complex vibration signals can be generated. However, in the prior art, the vibration signal of the bearing is directly used as the model to be input, so that the model is difficult to extract the characteristic information of the bearing fault from the complex vibration signal, and the accuracy of the bearing fault diagnosis is poor. In addition, the vibration signals have strong dependence on the time dimension, but in the existing method for predicting the bearing faults through the artificial neural network, the long-term dependence in the time sequence of the bearing vibration signals is difficult to effectively capture, and therefore the bearing fault diagnosis effectiveness is poor. Therefore, how to improve the accuracy and the effectiveness of bearing fault diagnosis is a technical problem to be solved.
Disclosure of Invention
Aiming at the defects in the prior art, the invention aims to solve the technical problems that: how to provide a bearing fault diagnosis method based on SDP and visual transducer codes, through effectively fusing SDP algorithm and visual transducer codes, the characteristic information of bearing faults can be effectively extracted from complex vibration signals, and meanwhile, long-term dependency relationship in the time sequence of the bearing vibration signals can be better captured, so that the bearing fault diagnosis accuracy and effectiveness are improved, and a new thought is provided for bearing fault diagnosis.
In order to solve the technical problems, the invention adopts the following technical scheme:
a bearing fault diagnosis method based on SDP and visual transducer coding, comprising:
s1: acquiring a vibration signal of a bearing to be diagnosed;
s2: converting the vibration signal of the bearing to be diagnosed into a corresponding SDP diagram through an SDP algorithm;
s3: inputting the SDP diagram of the bearing to be diagnosed into a trained fault diagnosis model, and outputting to obtain a corresponding fault type predicted value;
the fault diagnosis model comprises the following processing steps:
s301: dividing an input SDP image into a series of SDP image block sequences;
s302: inputting the SDP image block sequence into an embedding layer to linearly add position embedding in the vector of the SDP image block sequence to obtain a corresponding position embedded image block sequence;
s303: inputting the position embedded image block sequence into a visual transducer encoder to capture local information and global information in the SDP image block sequence and learn the dependency relationship between different parts in the SDP image block sequence to obtain a corresponding SDP image block sequence coding representation;
s304: coding SDP image block sequences to represent the input MLP header layer, obtaining probability values of various fault types, and further taking the fault type corresponding to the maximum probability value as a fault type prediction value;
s4: and taking the output fault type predicted value as a fault diagnosis result of the bearing to be diagnosed.
Preferably, the SDP graph is generated by:
s201: acquiring a vibration time sequence signal of a bearing;
s202: preprocessing the vibration time sequence signal;
s203: segmenting the preprocessed vibration time sequence signals to obtain a plurality of segmented signals; wherein each segment signal is representative of a characteristic of the vibration timing signal over a time window;
s204: and executing SDP operation on each segmented signal to obtain a corresponding SDP diagram.
Preferably, when performing the SDP operation, an arbitrary point in the segmented signal is converted into a point in the polar coordinate space, so as to form an SDP graph under polar coordinates;
specific:
in the timing signal N, x i For the amplitude of the timing signal N at the ith point in time, x i+a For time-domain signals x with hysteresis coefficients i A corresponding amplitude value;
when a webValue x i Mapped to polar coordinate space P (r (i), Θ (i), phi (i)), polar coordinate radius r (i) is defined by amplitude x i Mapping, the formula is described as:
wherein: x is x max 、x min Respectively representing the maximum value and the minimum value of the time domain signal N;
the rotation angles along the initial line in the polar coordinate space are respectively Θ (i) and phi (i), which are defined by the sum of x and Θ (i) i Adjacent amplitude x i+a Mapping results, the formula is described as:
wherein: theta (i) and phi (i) represent the time domain point x i+a Clockwise and counterclockwise rotation angles; x is x i+a Representing a time domain signal x with hysteresis coefficients i A corresponding amplitude value; zeta and a represent adjustment amplification factors and time lag factors, and any point in the time domain signal can be intuitively represented as polar coordinates by adjusting the amplification factors zeta and the time lag factors a; wherein the SDP diagram is at an angle θ of the plane of mirror symmetry of one l Fan-shaped leaf images formed for symmetry axes.
Preferably, for an input SDP diagram x ε R h×w×c Where h denotes the height of the SDP diagram, w denotes the width of the SDP diagram, c denotes the number of channels of the SDP diagram, and is divided into N image blocks x of length and width p and channel number c i ∈R p*p*c 。
Preferably, N image blocks x i ∈R p*p*c Tiled as a one-dimensional sequence x p ∈R N×(p*p*c) Then the one-dimensional sequence x 'is processed through the linear projection layer' p ∈R N×D And performing linear projection, reserving position information, one-dimensional feature vectors and class labels of each image block, and mapping each image block into a D-dimensional vector space to obtain a corresponding position embedded image block sequence.
Preferably, the visual transducer encoder is formed by stacking a plurality of identical module layers, wherein each module layer comprises two sub-layers, and each sub-layer consists of a multi-head self-attention layer and an MLP feedforward network;
in the multi-head self-attention layer, the input positions are embedded into the image block sequence through a plurality of attention heads, wherein each attention head corresponds to different attention points in the learning SDP image block sequence, each attention head generates an output, and the outputs of all the attention heads are combined in a final result; in the MLP feed forward network, feature extraction and nonlinear mapping are performed on the final result of the multi-head self-attention layer output.
Preferably, in the visual transducer encoder, layer normalization is used at the end of each sub-layer, the output of each sub-layer being expressed as:
o＝LayerNorm(x+Sublayer(x))；
wherein: layerNorm (·) represents the normalization function; sublayer (x) represents a function of the multi-headed self-attention layer or MLP feed-forward network in each sub-layer.
Preferably, the multi-head self-attention layer performs joint attention on the characteristics of different representing subspaces at different positions in the position embedded image block sequence through a multi-head self-attention mechanism; the self-attention mechanism calculates the attention weight of the feature matrix by adopting the scale dot product attention;
assume that the Q, K dimension of the input is d K V dimension d V Then the point multiplication operation for Q and each K is calculated and divided byThen, calculating the attention weight by applying a Softmax function;
the formula is described as follows:
wherein: attention (Q, K, V) represents an Attention weight; q represents a query matrix; k represents a key matrix; v represents a matrix of values; q, K, V are respectively composed of input characteristic matrix X f And parameter matrix W Q 、W K 、W V Multiplication.
Preferably, the internal structure of the MLP feedforward network comprises a full connection layer, a GELU activation function layer and a dropout function layer;
the output of the GELU activation function layer is as follows:
wherein: x represents the input of the GELU activation function layer; erf (. Cndot.) represents a Gaussian error function.
Preferably, the MLP header layer comprises a full-connection layer and a GELU activation function layer which are sequentially connected in series, the SDP image block sequence processed by the visual transducer encoder is encoded and input into the MLP header layer to obtain probability values of various fault types, and then the fault type corresponding to the maximum probability value is used as a fault type prediction value.
Compared with the prior art, the bearing fault diagnosis method based on SDP and visual transducer codes has the following beneficial effects:
according to the invention, the vibration signal of the bearing to be diagnosed is converted into an SDP diagram through an SDP algorithm, and then the fault type is predicted through a fault diagnosis model. On one hand, the invention converts the vibration signal into the SDP diagram through the SDP algorithm, wherein the SDP algorithm utilizes a mode of synchronously modulating and demodulating the vibration signal to extract the modulation component related to the fault, so that the characteristic information of the bearing fault can be effectively extracted from the complex vibration signal, and the accuracy of bearing fault diagnosis is further ensured; meanwhile, the SDP algorithm can extract a plurality of characteristic frequencies at the same time, so that the condition of bearing faults can be more comprehensively revealed, and compared with a traditional spectrum analysis method, more accurate and detailed fault characteristic information can be provided, and the accurate diagnosis of bearing faults can be more easily realized; in addition, the SDP algorithm has stronger noise suppression performance, and can effectively reduce the interference of noise on fault characteristic frequency, thereby further improving the reliability of fault diagnosis.
On the other hand, in the fault diagnosis model, the SDP diagram is firstly divided into the image block sequences to more clearly represent the characteristics in the bearing vibration signals, and each image block can capture the local details of the vibration signals, so that the subsequent characteristic extraction and diagnosis are more accurate. The SDP image block sequence is subjected to position embedding so as to ensure that the position information of each image block is not lost when the feature extraction and diagnosis are performed, and the result of the image block can be conveniently mapped back into the original vibration signal, so that a more visual diagnosis result is provided; meanwhile, the robustness to noise and interference can be improved through position embedding, and the sensitivity to noise and interference can be reduced through using the information embedded by the position in the diagnosis process, so that the accuracy and reliability of bearing fault diagnosis are improved; in addition, each image block can be processed in parallel through position embedding, the characteristic extraction and diagnosis processes can be accelerated by utilizing the advantages of parallel calculation, and the efficiency and practicability of bearing fault diagnosis are improved. Finally, capturing local information and global information in the SDP image block sequence through a visual transducer encoder and learning the dependency relationship between different parts in the SDP image block sequence, wherein the visual transducer encoder can better capture the long-term dependency relationship in the time sequence of the bearing vibration signal compared with a deep learning network such as a cyclic neural network and the like, thereby improving the effectiveness of bearing fault diagnosis; meanwhile, the visual transducer encoder can process different parts of the input SDP image block sequence in parallel without sequential processing, is more efficient for processing longer sequences and a large amount of data, and does not increase training and reasoning time, thereby further improving the efficiency of bearing fault diagnosis.
Drawings
For the purpose of making the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in further detail with reference to the accompanying drawings, in which:
FIG. 1 is a logic block diagram of a bearing fault diagnosis method based on SDP and visual transducer coding;
fig. 2 is a schematic diagram of an SDP algorithm conversion SDP diagram;
FIG. 3 is a network block diagram of a visual transducer encoder;
fig. 4 is a network structure diagram of each sub-layer in the visual transducer encoder.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention more clear, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. It will be apparent that the described embodiments are some, but not all, embodiments of the invention. The components of the embodiments of the present invention generally described and illustrated in the figures herein can be arranged and designed in a wide variety of different configurations. Thus, the following detailed description of the embodiments of the invention, as presented in the figures, is not intended to limit the scope of the invention, as claimed, but is merely representative of selected embodiments of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
It should be noted that: like reference numerals and letters denote like items in the following figures, and thus once an item is defined in one figure, no further definition or explanation thereof is necessary in the following figures. In the description of the present invention, it should be noted that, directions or positional relationships indicated by terms such as "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., are directions or positional relationships based on those shown in the drawings, or are directions or positional relationships conventionally put in use of the inventive product, are merely for convenience of describing the present invention and simplifying the description, and are not indicative or implying that the apparatus or element to be referred to must have a specific direction, be constructed and operated in a specific direction, and thus should not be construed as limiting the present invention. Furthermore, the terms "first," "second," "third," and the like are used merely to distinguish between descriptions and should not be construed as indicating or implying relative importance. Furthermore, the terms "horizontal," "vertical," and the like do not denote a requirement that the component be absolutely horizontal or overhang, but rather may be slightly inclined. For example, "horizontal" merely means that its direction is more horizontal than "vertical" and does not mean that the structure must be perfectly horizontal, but may be slightly tilted. In the description of the present invention, it should also be noted that, unless explicitly specified and limited otherwise, the terms "disposed," "mounted," "connected," and "connected" are to be construed broadly, and may be, for example, fixedly connected, detachably connected, or integrally connected; can be mechanically or electrically connected; can be directly connected or indirectly connected through an intermediate medium, and can be communication between two elements. The specific meaning of the above terms in the present invention will be understood in specific cases by those of ordinary skill in the art.
The following is a further detailed description of the embodiments:
examples:
the embodiment discloses a bearing fault diagnosis method based on SDP and visual transducer codes.
As shown in fig. 1, the bearing fault diagnosis method based on SDP and visual transducer coding includes:
s1: acquiring a vibration signal of a bearing to be diagnosed;
s2: converting the vibration signal of the bearing to be diagnosed into a corresponding SDP diagram through an SDP (Symmetrized Dot Pattern, point symmetry mode) algorithm;
s3: inputting the SDP diagram of the bearing to be diagnosed into a trained fault diagnosis model, and outputting to obtain a corresponding fault type predicted value;
the fault diagnosis model comprises the following processing steps:
s301: dividing an input SDP image into a series of SDP image block sequences;
s302: inputting the SDP image block sequence into a linear projection layer serving as an embedding layer to linearly add position embedding into a vector of the SDP image block sequence to obtain a corresponding position embedded image block sequence;
s303: inputting the position embedded image block sequence into a visual transducer composed of a multi-head attention layer and a multi-layer perceptron layer to capture local information and global information in the SDP image block sequence and learn the dependency relationship between different parts in the SDP image block sequence to obtain a corresponding SDP image block sequence coding representation;
s304: coding SDP image block sequences to represent the input MLP header layer, obtaining probability values of various fault types, and further taking the fault type corresponding to the maximum probability value as a fault type prediction value;
in this embodiment, the MLP header layer includes a full connection layer and a GELU activation function layer sequentially connected in series, and the SDP image block sequence processed by the visual transducer encoder is encoded and input into the MLP header layer to obtain probability values of various fault types, and then the fault type corresponding to the maximum probability value is used as a fault type prediction value.
When the fault diagnosis model is trained, SDP graphs of different categories are used as new data sets and are divided into a training set and a testing set according to the ratio of 0.8:0.2.
S4: and taking the output fault type predicted value as a fault diagnosis result of the bearing to be diagnosed.
According to the invention, the vibration signal of the bearing to be diagnosed is converted into an SDP diagram through an SDP algorithm, and then the fault type is predicted through a fault diagnosis model. On one hand, the invention converts the vibration signal into the SDP diagram through the SDP algorithm, wherein the SDP algorithm utilizes a mode of synchronously modulating and demodulating the vibration signal to extract the modulation component related to the fault, so that the characteristic information of the bearing fault can be effectively extracted from the complex vibration signal, and the accuracy of bearing fault diagnosis is further ensured; meanwhile, the SDP algorithm can extract a plurality of characteristic frequencies at the same time, so that the condition of bearing faults can be more comprehensively revealed, and compared with a traditional spectrum analysis method, more accurate and detailed fault characteristic information can be provided, and the accurate diagnosis of bearing faults can be more easily realized; in addition, the SDP algorithm has stronger noise suppression performance, and can effectively reduce the interference of noise on fault characteristic frequency, thereby further improving the reliability of fault diagnosis.
On the other hand, in the fault diagnosis model, the SDP diagram is firstly divided into the image block sequences to more clearly represent the characteristics in the bearing vibration signals, and each image block can capture the local details of the vibration signals, so that the subsequent characteristic extraction and diagnosis are more accurate. The SDP image block sequence is subjected to position embedding so as to ensure that the position information of each image block is not lost when the feature extraction and diagnosis are performed, and the result of the image block can be conveniently mapped back into the original vibration signal, so that a more visual diagnosis result is provided; meanwhile, the robustness to noise and interference can be improved through position embedding, and the sensitivity to noise and interference can be reduced through using the information embedded by the position in the diagnosis process, so that the accuracy and reliability of bearing fault diagnosis are improved; in addition, each image block can be processed in parallel through position embedding, the characteristic extraction and diagnosis processes can be accelerated by utilizing the advantages of parallel calculation, and the efficiency and practicability of bearing fault diagnosis are improved. Finally, capturing local information and global information in the SDP image block sequence through a visual transducer encoder and learning the dependency relationship between different parts in the SDP image block sequence, wherein the visual transducer encoder can better capture the long-term dependency relationship in the time sequence of the bearing vibration signal compared with a deep learning network such as a cyclic neural network and the like, thereby improving the effectiveness of bearing fault diagnosis; meanwhile, the visual transducer encoder can process different parts of the input SDP image block sequence in parallel without sequential processing, is more efficient for processing longer sequences and a large amount of data, and does not increase training and reasoning time, thereby further improving the efficiency of bearing fault diagnosis.
In summary, the SDP algorithm and the visual transducer code are effectively fused, so that the characteristic information of the bearing fault can be effectively extracted from the complex vibration signals, and meanwhile, the long-term dependency relationship in the time sequence of the bearing vibration signals can be better captured, thereby improving the accuracy, the effectiveness and the efficiency of bearing fault diagnosis and providing a new thought for bearing fault diagnosis.
In a specific implementation process, the SDP algorithm adopted by the invention is a method commonly used for vibration signal processing and fault diagnosis, and the characteristics are extracted and the fault diagnosis is carried out by converting the vibration signal into an SDP image. Based on the characteristics of the SDP image, the fault type of the bearing can be diagnosed by feature extraction of the image.
As shown in connection with fig. 2, an SDP graph is generated by:
s201: acquiring a vibration time sequence signal (namely a time sequence signal of the vibration signal) of the bearing;
in this embodiment, the timing signals include signals in a normal operation state and signals in various fault states, such as damage to an inner ring, damage to an outer ring, damage to a ball, and the like.
S202: preprocessing the vibration time sequence signal;
in this embodiment, preprocessing includes denoising, filtering, normalization, and the like to ensure the quality and reliability of the signal.
S203: segmenting the preprocessed vibration time sequence signals to obtain a plurality of segmented signals; wherein each segment signal is representative of a characteristic of the vibration timing signal over a time window;
s204: and executing SDP operation on each segmented signal to obtain a corresponding SDP diagram.
In this embodiment, the segmented signal may be converted into the corresponding SDP graph by:
1) Calculating the polar radius of each time domain point of the segmented signal;
2) Calculating the clockwise and counterclockwise rotation angles of each time domain point of the segmented signal;
3) Mapping all time domain points onto a polar graph according to 1) and 2);
4) Rotating the drawn polar coordinate graph for 6 times according to 60 degrees to obtain a snowflake pattern similar to symmetry, namely an SDP graph;
5) Cutting the required image size according to the polar coordinate center position;
6) The segmented signals of all fault classes are looped 1) to 5) to generate an SDP image dataset.
Specifically, when the SDP operation is executed, any point in the segmented signal is converted into a point in a polar coordinate space, so as to form an SDP diagram under the polar coordinate;
in the timing signal N, x i For the amplitude of the timing signal N at the ith point in time, x i+a For time-domain signals x with hysteresis coefficients i A corresponding amplitude value;
when an amplitude x i Mapped to polar coordinate space P (r (i), Θ (i), phi (i)), polar coordinate radius r (i) is defined by amplitude x i Mapping, the formula is described as:
wherein: x is x max 、x min Respectively representing the maximum value and the minimum value of the time domain signal N;
the rotation angles along the initial line in the polar coordinate space are respectively Θ (i) and phi (i), which are defined by the sum of x and Θ (i) i Adjacent amplitude x i+a Mapping results, the formula is described as:
wherein: theta (i) and phi (i) represent the time domain point x i+a Clockwise and counterclockwise rotation angles; x is x i+a Representing a time domain signal x with hysteresis coefficients i A corresponding amplitude value; zeta and a represent the adjustment of the amplification factor and the time lag factor, and any point in the time domain signal can be realized by adjusting the amplification factor zeta and the time lag factor aIntuitively expressed as polar coordinates; where the SDP plot is the angle θ with the plane of symmetry of the L mirrors (l=0, 1,2,.. l Fan-shaped leaf images formed for symmetry axes.
In the invention, the SDP algorithm is adopted to extract the modulation component related to the fault by using the mode of synchronously modulating and demodulating the vibration signal, so that the characteristic information of the bearing fault can be effectively extracted from the complex vibration signal, and the accuracy of bearing fault diagnosis is further ensured; meanwhile, the SDP algorithm can extract a plurality of characteristic frequencies at the same time, so that the condition of bearing faults can be more comprehensively revealed, and compared with a traditional spectrum analysis method, more accurate and detailed fault characteristic information can be provided, and the accurate diagnosis of bearing faults can be more easily realized; in addition, the SDP algorithm has stronger noise suppression performance, and can effectively reduce the interference of noise on fault characteristic frequency, thereby further improving the reliability of fault diagnosis.
The SDP image adopted by the invention visualizes the frequency spectrum information of the vibration signal, so that the characteristics are more obvious and visual, and the fault diagnosis and analysis are convenient; meanwhile, the SDP image removes some high-frequency noise of the vibration signal, so that the characteristics are more obvious, and the accuracy and the robustness of diagnosis can be improved; in addition, the SDP images show different characteristics for different fault types, and the fault types of the bearings can be accurately judged through analysis of the characteristics.
In the implementation process, the linear projection layer is used as an embedded layer of the fault diagnosis model.
In the linear projection layer, for an input SDP graph x ε R h×w×c Where h denotes the height of the SDP diagram, w denotes the width of the SDP diagram, c denotes the number of channels of the SDP diagram, and is divided into N image blocks x of length and width p and channel number c i ∈R p*p*c . Simultaneously dividing N image blocks x i ∈R p*p*c Tiled as a one-dimensional sequence x p ∈R N×(p*p*c) Then the one-dimensional sequence x 'is processed through the linear projection layer' p ∈R N×D Performing linear projection, retaining position information, one-dimensional feature vectors and class labels of each image block, and mapping each image block into a D-dimensional vector spaceAnd obtaining a corresponding position embedded image block sequence.
The invention performs position embedding on the vectors of the SDP image block sequence through the linear projection layer. The linear projection layer can perform dimension reduction processing on the SDP image block sequence, map high-dimension data into a low-dimension space, be beneficial to highlighting the characteristics of fault types, improve the identification capability of the characteristics, and convert the SDP image block sequence into characteristic vectors with more discrimination so that the fault types are more obvious; meanwhile, the linear projection layer can reduce the data dimension to a proper range, and can accelerate the fault diagnosis process and improve the bearing fault diagnosis efficiency. In addition, the unknown embedding of the linear projection layer enables the model to be better suitable for different SDP image block sequences, and the generalization capability of the model is improved.
Referring to fig. 3, the visual transducer of the fault diagnosis model of the present invention is composed of a multi-head attention layer and a multi-layer sensor layer.
Specifically, the visual transducer encoder is formed by stacking a plurality of identical module layers, wherein each module layer comprises two sub-layers, and each sub-layer consists of a multi-head self-attention layer and an MLP feedforward network; in the multi-head self-attention layer, the input positions are embedded into the image block sequence through a plurality of attention heads, wherein each attention head corresponds to different attention points in the learning SDP image block sequence, each attention head generates an output, and the outputs of all the attention heads are combined in a final result; in the MLP feed forward network, feature extraction and nonlinear mapping are performed on the final result of the multi-head self-attention layer output.
Referring to fig. 4, in the visual transducer encoder, layer normalization is used at the end of each sub-layer, and the output of each sub-layer is expressed as:
o＝LayerNorm(x+Sublayer(x))；
wherein: layerNorm (·) represents the normalization function; sublayer (x) represents a function of the multi-headed self-attention layer or MLP feed-forward network in each sub-layer.
The visual transducer of the present invention is composed of a multi-headed attention layer and a multi-layered sensor layer. The multi-head attention layer allows the vision transducer to learn attention points at different positions in the sequence in different attention heads, so that the vision transducer can learn correlations of different fault modes at different frequencies of the vibration signal through a multi-head attention mechanism, and can understand and distinguish different fault modes, thereby better capturing long-term dependence of the bearing vibration signal in the time sequence. While the multi-layer perceptron layer provides additional nonlinear modeling capability, more layers can be added to the visual transducer encoder to increase the representation capability of the model, which is particularly important for bearing failure diagnosis tasks, because the characteristics of the vibration signal tend to be nonlinear, requiring a model with sufficient expressive power to capture these characteristics, as well as to assist in achieving capture of long-term dependencies.
In the implementation process, multi-head Self-Attention (Multi-head Self-Attention) is an Attention mechanism for sequence modeling and natural language processing tasks, and is one of key components in a transducer model. A self-attention mechanism means that each element in a sequence can be associated with other elements in the sequence and the associations weighted. Multi-head self-attention expands the attention mechanism and models through a plurality of independent attention heads.
In multi-head self-attention, the input sequence is processed through multiple attention heads, respectively. Each attention head learns a different attention weight. This allows the model to focus on different semantic information simultaneously, capturing multiple different aspects of the input sequence. Specifically, the multi-headed self-attention operates as follows:
1) Each element in the input sequence is mapped to a vector space of different queries, keys and values by linear transformation;
2) Each attention head obtains attention weight by calculating query-key similarity;
3) Calculating a weighted value for each attention header, the value being a dot product of the attention weight and the corresponding value;
4) The weighted values of the plurality of attention heads are spliced or combined together and the final output is obtained by a linear transformation.
Through multi-headed self-attention, the visual transducer can model the input sequence from different angles, capturing different semantic information.
Specifically, the multi-head self-attention layer carries out joint attention on the characteristics of different representing subspaces at different positions in the position embedded image block sequence through a multi-head self-attention mechanism; the self-attention mechanism calculates the attention weight of the feature matrix by adopting the scale dot product attention, and calculates dot product attention according to the proportion;
it should be noted that, dot product attention is a common method for calculating attention weight, and is a specific mathematical formula for calculating attention weight. Attention weights are calculated actual values representing the degree of attention, which are values used in the attention mechanism to measure the degree of attention of different elements.
Assume that the Q, K dimension of the input is d K V dimension d V Then the point multiplication operation for Q and each K is calculated and divided byThen, calculating the attention weight by applying a Softmax function;
the formula is described as follows:
wherein: attention (Q, K, V) represents an Attention weight; q represents a query matrix; k represents a key matrix; v represents a matrix of values; q, K, V are respectively composed of input characteristic matrix X f And parameter matrix W Q 、W K 、W V Multiplication.
In particular, in the visual transducer encoder, an MLP feed Forward Network (Multi-Layer Perceptron Feed-Forward Network) is a key component for extracting features and performing nonlinear transformation. The MLP feed forward network is one of the sub-modules of each encoder layer in the transducer model. The function of the method is to further process the characteristics obtained after self-attention mechanism so as to enhance the representation capability and expression capability of the characteristics. The MLP feed forward network consists of two linear transformation layers and an activation function. The linear transformation layer performs linear transformation and mapping on the input features, and complexity and expression capability of the features are increased. The activation function then introduces a nonlinear transformation that further enriches the representational capabilities of the feature.
In each encoder layer, the input to the MLP feed forward network is a feature derived by a self-attention mechanism. The feature may pass through a fully connected hidden layer, where the hidden layer is typically larger in dimension than the input feature. The output of the hidden layer then passes through a fully connected output layer, remapping the feature dimensions back to the original dimensions. Finally, an activation function (e.g., reLU) applies a nonlinear transformation to the output features. Through the MLP feed forward network, the encoder is able to perform more complex feature extraction and nonlinear transformation. This helps to improve the representational capacity of the features so that the model better captures semantic information and complex relationships in the input sequence.
Specifically, the internal structure of the MLP feedforward network comprises a full connection layer, a GELU activation function layer and a dropout function layer;
to improve network convergence, a gaussian error linear unit (gel) activation function is used in place of the original gel activation function in the MLP feed forward network. The output of the GELU activation function layer is as follows:
wherein: x represents the input of the GELU activation function layer; erf (. Cndot.) represents a Gaussian error function.
Finally, it should be noted that the above embodiments are only for illustrating the technical solution of the present invention and not for limiting the technical solution, and those skilled in the art should understand that modifications and equivalents may be made to the technical solution of the present invention without departing from the spirit and scope of the present invention, and all such modifications and equivalents are included in the scope of the claims.
Claims (10)
1. A bearing fault diagnosis method based on SDP and visual transducer coding, comprising:
s1: acquiring a vibration signal of a bearing to be diagnosed;
s2: converting the vibration signal of the bearing to be diagnosed into a corresponding SDP diagram through an SDP algorithm;
s3: inputting the SDP diagram of the bearing to be diagnosed into a trained fault diagnosis model, and outputting to obtain a corresponding fault type predicted value;
the fault diagnosis model comprises the following processing steps:
s301: dividing an input SDP image into a series of SDP image block sequences;
s302: inputting the SDP image block sequence into an embedding layer to linearly add position embedding in the vector of the SDP image block sequence to obtain a corresponding position embedded image block sequence;
s303: inputting the position embedded image block sequence into a visual transducer encoder to capture local information and global information in the SDP image block sequence and learn the dependency relationship between different parts in the SDP image block sequence to obtain a corresponding SDP image block sequence coding representation;
s304: coding SDP image block sequences to represent the input MLP header layer, obtaining probability values of various fault types, and further taking the fault type corresponding to the maximum probability value as a fault type prediction value;
s4: and taking the output fault type predicted value as a fault diagnosis result of the bearing to be diagnosed.
2. The bearing fault diagnosis method based on SDP and visual transducer coding as claimed in claim 1, wherein in step S2, an SDP graph is generated by:
s201: acquiring a vibration time sequence signal of a bearing;
s202: preprocessing the vibration time sequence signal;
s203: segmenting the preprocessed vibration time sequence signals to obtain a plurality of segmented signals; wherein each segment signal is representative of a characteristic of the vibration timing signal over a time window;
s204: and executing SDP operation on each segmented signal to obtain a corresponding SDP diagram.
3. The bearing fault diagnosis method based on SDP and visual transducer coding as claimed in claim 2, wherein: in step S204, when performing the SDP operation, an arbitrary point in the segmented signal is converted into a point in the polar coordinate space, so as to form an SDP graph under the polar coordinate;
specific:
in the timing signal N, x i For the amplitude of the timing signal N at the ith point in time, x i+a For time-domain signals x with hysteresis coefficients i A corresponding amplitude value;
when an amplitude x i Mapped to polar coordinate space P (r (i), Θ (i), phi (i)), polar coordinate radius r (i) is defined by amplitude x i Mapping, the formula is described as:
wherein: x is x max 、x min Respectively representing the maximum value and the minimum value of the time domain signal N;
the rotation angles along the initial line in the polar coordinate space are respectively Θ (i) and phi (i), which are defined by the sum of x and Θ (i) i Adjacent amplitude x i+a Mapping results, the formula is described as:
wherein: theta (i) and phi (i) represent the time domain point x i+a Clockwise and counterclockwise rotation angles; x is x i+a Representing a time domain signal x with hysteresis coefficients i A corresponding amplitude value; zeta and a represent adjustment amplification factors and time lag factors, and any point in the time domain signal can be intuitively represented as polar coordinates by adjusting the amplification factors zeta and the time lag factors a; wherein the SDP diagram is at an angle θ of the plane of mirror symmetry of one l Fan-shaped leaf images formed for symmetry axes.
4. The bearing fault diagnosis method based on SDP and visual transducer coding as claimed in claim 1, wherein: in step S301, for the input SDP diagram x εR h×w×c Where h denotes the height of the SDP diagram, w denotes the width of the SDP diagram, c denotes the number of channels of the SDP diagram, and is divided into N image blocks x of length and width p and channel number c i ∈R p*p*c 。
5. The method for diagnosing a bearing failure based on SDP and visual transducer coding as recited in claim 4, wherein: in step S302, N image blocks x i ∈R p*p*c Tiled as a one-dimensional sequence x p ∈R N×(p*p*c) Then the one-dimensional sequence x 'is processed through the linear projection layer' p ∈R N×D And performing linear projection, reserving position information, one-dimensional feature vectors and class labels of each image block, and mapping each image block into a D-dimensional vector space to obtain a corresponding position embedded image block sequence.
6. The bearing fault diagnosis method based on SDP and visual transducer coding as claimed in claim 1, wherein: in step S303, the visual transducer encoder is formed by stacking a plurality of identical module layers, wherein each module layer comprises two sub-layers, and each sub-layer is respectively composed of a multi-head self-attention layer and an MLP feedforward network;
in the multi-head self-attention layer, the input positions are embedded into the image block sequence through a plurality of attention heads, wherein each attention head corresponds to different attention points in the learning SDP image block sequence, each attention head generates an output, and the outputs of all the attention heads are combined in a final result; in the MLP feed forward network, feature extraction and nonlinear mapping are performed on the final result of the multi-head self-attention layer output.
7. The SDP and visual transducer code based bearing fault diagnosis method as claimed in claim 6, wherein: in the visual transducer encoder, layer normalization is employed at the end of each sub-layer;
the output of each sub-layer is expressed as:
o＝LayerNorm(x+Sublayer(x))；
wherein: layerNorm (·) represents the normalization function; sublayer (x) represents a function of the multi-headed self-attention layer or MLP feed-forward network in each sub-layer.
8. The SDP and visual transducer code based bearing fault diagnosis method as claimed in claim 6, wherein: the multi-head self-attention layer carries out joint attention on the characteristics of different representing subspaces at different positions in the position embedded image block sequence through a multi-head self-attention mechanism; the self-attention mechanism calculates the attention weight of the feature matrix by adopting the scale dot product attention;
assume that the Q, K dimension of the input is d K V dimension d V Then the point multiplication operation for Q and each K is calculated and divided byThen, calculating the attention weight by applying a Softmax function;
the formula is described as follows:
wherein: attention (Q, K, V) represents an Attention weight; q represents a queryA matrix; k represents a key matrix; v represents a matrix of values; q, K, V are respectively composed of input characteristic matrix X f And parameter matrix W Q 、W K 、W V Multiplication.
9. The SDP and visual transducer code based bearing fault diagnosis method as claimed in claim 6, wherein: the internal structure of the MLP feedforward network comprises a full connection layer, a GELU activation function layer and a dropout function layer;
the output of the GELU activation function layer is as follows:
wherein: x represents the input of the GELU activation function layer; erf (. Cndot.) represents a Gaussian error function.
10. The bearing fault diagnosis method based on SDP and visual transducer coding as claimed in claim 1, wherein: in step S304, the MLP header layer includes a full connection layer and a GELU activation function layer sequentially connected in series, and the SDP image block sequence processed by the visual transducer encoder is encoded and input into the MLP header layer to obtain probability values of various fault types, and then the fault type corresponding to the maximum probability value is used as a fault type prediction value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311106774.8A CN117009770A (en) | 2023-08-30 | 2023-08-30 | Bearing fault diagnosis method based on SDP and visual transducer codes |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311106774.8A CN117009770A (en) | 2023-08-30 | 2023-08-30 | Bearing fault diagnosis method based on SDP and visual transducer codes |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117009770A true CN117009770A (en) | 2023-11-07 |
Family
ID=88569086
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311106774.8A Pending CN117009770A (en) | 2023-08-30 | 2023-08-30 | Bearing fault diagnosis method based on SDP and visual transducer codes |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117009770A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117520993A (en) * | 2023-12-29 | 2024-02-06 | 杭州杭叉桥箱有限公司 | Electric drive axle test system and method thereof |
| CN117592571A (en) * | 2023-12-05 | 2024-02-23 | 武汉华康世纪医疗股份有限公司 | Air conditioning unit fault type diagnosis method and system based on big data |
| CN117830750A (en) * | 2024-03-04 | 2024-04-05 | 青岛大学 | Mechanical fault prediction method based on graph converter |
| CN117830750B (en) * | 2024-03-04 | 2024-06-04 | 青岛大学 | Mechanical fault prediction method based on graph converter |
-
2023
- 2023-08-30 CN CN202311106774.8A patent/CN117009770A/en active Pending
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117592571A (en) * | 2023-12-05 | 2024-02-23 | 武汉华康世纪医疗股份有限公司 | Air conditioning unit fault type diagnosis method and system based on big data |
| CN117592571B (en) * | 2023-12-05 | 2024-05-17 | 武汉华康世纪医疗股份有限公司 | Air conditioning unit fault type diagnosis method and system based on big data |
| CN117520993A (en) * | 2023-12-29 | 2024-02-06 | 杭州杭叉桥箱有限公司 | Electric drive axle test system and method thereof |
| CN117520993B (en) * | 2023-12-29 | 2024-03-08 | 杭州杭叉桥箱有限公司 | Electric drive axle test system and method thereof |
| CN117830750A (en) * | 2024-03-04 | 2024-04-05 | 青岛大学 | Mechanical fault prediction method based on graph converter |
| CN117830750B (en) * | 2024-03-04 | 2024-06-04 | 青岛大学 | Mechanical fault prediction method based on graph converter |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Chen et al. | A deep learning method for bearing fault diagnosis based on cyclic spectral coherence and convolutional neural networks | |
| Guo et al. | Intelligent fault diagnosis method based on full 1-D convolutional generative adversarial network | |
| CN117009770A (en) | Bearing fault diagnosis method based on SDP and visual transducer codes | |
| Lang et al. | Artificial intelligence-based technique for fault detection and diagnosis of EV motors: A review | |
| WO2022077901A1 (en) | Bearing failure mode diagnosis method using small sample data sets, and system | |
| Zhang et al. | DeepHealth: A self-attention based method for instant intelligent predictive maintenance in industrial Internet of Things | |
| Basu et al. | Use of artificial neural network in pattern recognition | |
| CN111582320B (en) | Dynamic individual identification method based on semi-supervised learning | |
| CN113780412B (en) | Fault diagnosis model training method and system and fault diagnosis method and system | |
| CN114818774A (en) | Intelligent gearbox fault diagnosis method based on multi-channel self-calibration convolutional neural network | |
| CN111860446A (en) | Detection system and method for unknown mode of satellite remote measurement time sequence data | |
| Azzalini et al. | A minimally supervised approach based on variational autoencoders for anomaly detection in autonomous robots | |
| CN114913396A (en) | Motor bearing fault diagnosis method | |
| CN114004135A (en) | Agricultural machinery bearing fault type diagnosis method and system based on Transformer neural network | |
| CN111795819B (en) | Gear box fault diagnosis method integrating vibration and current signal collaborative learning | |
| CN116011507A (en) | Rare fault diagnosis method for fusion element learning and graph neural network | |
| Chen et al. | Discovering state variables hidden in experimental data | |
| Wang et al. | Ensemble clustering-based fault diagnosis method incorporating traditional and deep representation features | |
| Wang et al. | C-ECAFormer: A new lightweight fault diagnosis framework towards heavy noise and small samples | |
| Wang et al. | Image deep learning in fault diagnosis of mechanical equipment | |
| Yu et al. | Deep transfer learning rolling bearing fault diagnosis method based on convolutional neural network feature fusion | |
| CN113530850A (en) | Centrifugal pump fault diagnosis method based on EAS (electronic article surveillance) and stacked capsule self-encoder | |
| Lu et al. | Rotating Machinery Fault Diagnosis Under Multiple Working Conditions via A Time Series Transformer Enhanced by Convolutional Neural Network | |
| CN117763494A (en) | Bearing diagnosis method and system based on multi-mode and multi-scale fusion network | |
| CN116894207A (en) | Intelligent radiation source identification method based on Swin transducer and transfer learning |
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 |