Heavy machine transmission system life evaluation method based on sparse trait guide layer net

Abstract

The invention discloses a high-speed heavy machine transmission system service life evaluation method based on a sparse trait guide layer net. The method comprises the following steps of weakening deterioration data disturbance of a high-speed heavy machine transmission system; characterizing high-dimensional sparse characteristics of the service life of the transmission system of the high-speed heavy machine; determining the defect overflow time of the transmission system of the high-speed heavy machine; establishing a stack guide layer network model; and evaluating the service life of the transmission system of the high-speed heavy machine based on the stacked guide layer net. According to the method, key life characteristics of the transmission system can be efficiently mined from a multi-dimensional time sequence, and important dynamic characteristics are screened out, so that interference of redundant information is effectively reduced; short-term dynamic characteristics and long-term dependence characteristics in the equipment operation process can be accurately captured, and the stability and robustness of life evaluation are further improved; the life prediction precision under complex working conditions can be remarkably improved, and the method has important engineering application value and practical significance.

Description

Technical Field

[0001] The present invention relates to a system life assessment method, and in particular to a heavy machine transmission system life assessment method based on a sparse characteristic-guided layer network. Background Art

[0002] With the increasing complexity of industrial equipment and the diversity of operating environments, the life assessment of high-speed heavy machinery transmission systems is facing new challenges. Traditional life assessment methods are usually based on rule models or single feature analysis, which makes it difficult to capture the nonlinear dynamic relationship of multi-dimensional time series characteristics in the transmission system, especially under complex working conditions, where the sparse nature of equipment state changes cannot be fully explored. This limitation not only limits the accuracy of life prediction, but may also lead to the inability to provide early warning of potential risks in equipment operation. In this context, a new life assessment method for high-speed heavy machinery transmission systems is urgently needed to solve the difficulties of traditional methods in nonlinear dynamic modeling and provide a more accurate and reliable decision support tool for industrial equipment health management. Summary of the invention

[0003] Purpose of the invention: The purpose of the present invention is to provide a method for evaluating the life of a heavy machinery transmission system based on a sparse characteristic-guided layer network.

[0004] Technical solution: The method for evaluating the life of a high-speed heavy-duty transmission system based on a sparse characteristic-guided layer network of the present invention comprises the following steps:

[0005] (1) The deterioration data disturbance of high-speed heavy-duty machine transmission system is weakened;

[0006] (2) High-dimensional sparse characteristic representation of the life of high-speed heavy-duty transmission systems;

[0007] (3) Determination of the spillover time of defects in the transmission system of high-speed heavy machinery;

[0008] (4) Establishment of stacked guidance layer network model;

[0009] (5) Life assessment of high-speed heavy machinery transmission system based on stacked guidance layer network.

[0010] Furthermore, the step (1) includes first defining the optimization objective function and characterizing the relationship constraint, and then performing the modal function u of the original signal. k and the center frequency ω k Perform minimum optimization solution.

[0011] Furthermore, the definition of the optimization objective function and the characterization of the relational constraints include:

[0012] According to the running state of the high-speed bearing, the periodic motion and degradation characteristics of the bearing, the bearing degradation signal x(t) is obtained, and the bandwidth of the signal is smoothed using the Hanning window function to achieve uniform bandwidth mapping.

[0013] For a vibration signal x(t) of length n, an optimization objective function is constructed. The model is used to calculate the modal function u of the original signal. k and the center frequency ω k For minimal optimization, the optimization objective function is defined as:

[0014]

[0015] in Represents the partial derivative operation on time t; δ(t) is the unit impulse function, which is used to localize or filter the signal; is a singular kernel function; * represents a convolution operator; represents the sum of squares of the signal;

[0016] Establish an optimization constraint function to characterize the relationship between the original signal and the modal function of the original signal:

[0017]

[0018] x(t) represents a vibration signal of length n; u k is the mode function of the original signal.

[0019] Furthermore, the modal function u of the original signal k and the center frequency ω k The formulas for minimum optimization solutions are:

[0020]

[0021] in, represents the estimated value of the kth mode signal in the frequency domain ω at the n+1th iteration; Represents the estimated value of the input signal x(t) in the frequency domain ω; represents a constraint term; α represents the smoothing index.

[0022] Furthermore, the step (2) comprises:

[0023] According to the optimal modal signal u determined in step (1) k (ω),u k (ω) is more representative of the bearing life characteristics in the long-term trend and is more representative of the optimal modal signal u k (ω) is normalized to avoid the gradient explosion problem. Specifically, the normalized optimal modal signal for:

[0024]

[0025] in, represents the standard normalized signal, u k (ω) represents the vibration information of the high-speed bearing, max() represents the maximum value operation, and min() represents the minimum value operation.

[0026] Furthermore, the normalized optimal modal signal is processed by dimensionality enhancement through the Gram angle field to achieve a high-dimensional representation of the one-dimensional signal. The high-dimensional representation form of the one-dimensional signal is GADF i，j for:

[0027]

[0028] Among them, GADF is the Gram angular field transform, is the row standard normalized signal, is the column standard normalized signal; the signal after Gram angle field mapping is calculated according to GADF i，j The spectral characteristics of the multidimensional sparse characteristic Γ are characterized by:

[0029]

[0030] Among them, I is the length of the time series after mapping, J is the dimension of the time series after mapping, and GADF i，j It is a high-dimensional representation of a one-dimensional signal.

[0031] Furthermore, the step (3) comprises:

[0032] Based on the original bearing degradation signal x(t), the multi-dimensional sparse characteristics extracted in step (2) are rearranged in time sequence to obtain the time sequence high-dimensional sparse characteristics Γ(t), whose envelope energy E(t) can be expressed as:

[0033]

[0034] Among them, Γ(t) represents the time series high-dimensional sparse characteristics, t represents the time series information of the high-dimensional sparse characteristics, E(t) represents the envelope energy of the high-dimensional sparse characteristics, and T is the current moment information;

[0035] Weibull distribution is suitable for describing the failure behavior of mechanical equipment such as bearings. The probability density function that obeys the Weibull distribution is defined as:

[0036]

[0037] Among them, T is the current time information, β is the shape parameter, and η is the scale parameter. The optimal values ​​of the shape parameter β and the scale parameter η are determined by the linear regression method:

[0038] ln(-ln(S(T)))=βln(T)-βln(η)

[0039] The constructed probability density function can be used to derive the survival function S(T) that represents the probability of no damage occurring before time T and the risk function H(T) that represents the probability of instantaneous damage occurring at time T:

[0040]

[0041] Where T is the current moment information, β is the shape parameter, and η is the scale parameter;

[0042] Determining the survival probability threshold can lead to the defect spillover time T FFT :

[0043] T FFT =min{T|S(T)≤S th}

[0044] Among them, the survival probability threshold is 95%.

[0045] Furthermore, the step (4) comprises:

[0046] According to step (3), the high-dimensional sparse feature representation Γ(t) after the defect overflow time T FFT After confirmation, in order to deeply mine the deep sparse characteristic information and establish the guided layer network model, the guided layer network model takes the single-layer recursive layer network as the core, and constructs the core formula of the single-layer recursive layer network:

[0047] h t =σ(W h h t-1 +W x Γ(t)+b h )

[0048] Where Γ(t) represents the high-dimensional sparse nature of the time series, h t represents the temporal hidden layer, σ-dimensional nonlinear activation function, W h , W x 、b h Represents the weight and bias parameters of the network.

[0049] Furthermore, the establishment of the guidance layer network model includes:

[0050] The guided layer network is constructed by stacking multiple layers of recursive units in time sequence. The output of each layer of recursive units is used as the input of the next layer, thereby realizing multi-layer feature recognition. The formula is:

[0051]

[0052] Among them, W h , W x 、b h represents the weight and bias parameters of the network, are the weight and bias parameters of the lth layer.

[0053] For the constructed regression layer network, the regression task loss function is introduced Optimize and adjust parameters:

[0054]

[0055] Among them, Γ(t) represents the high-dimensional sparse characteristics of time series, represents the trait mean, Represents the regression task loss function.

[0056] Furthermore, the step (5) comprises:

[0057] According to the guided layer network model established in step (4), the life of the high-speed heavy machine transmission system is evaluated. The input high-dimensional sparse features are linearly transformed through the first fully connected layer of the guided layer network model to obtain the life evaluation result:

[0058] θ=g(h(Γ(t);θ h );θ g )

[0059] Among them, Θ is the life assessment result of the high-speed heavy machine transmission system, g() represents the activation function, h() represents the hidden layer operation, and θ is the interference function.

[0060] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: the present invention combines sparse feature extraction and guidance layer network modeling technology, which can efficiently mine the key life characteristics of the transmission system from the multi-dimensional time series, and through the sparse feature enhancement mechanism, it can screen out important dynamic features, thereby effectively reducing the interference of redundant information; at the same time, through the multi-layer dynamic recursive modeling of the guidance layer network, it can accurately capture the short-term dynamic characteristics and long-term dependent characteristics during the operation of the equipment, further improving the stability and robustness of the life assessment; it can significantly improve the life prediction accuracy under complex working conditions, and has important engineering application value and practical significance. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0062] The technical solution of the present invention is further described below in conjunction with the accompanying drawings.

[0063] like Figure 1 As shown, the present invention comprises the following steps:

[0064] (1) The deterioration data disturbance of high-speed heavy-duty machine transmission system is weakened;

[0065] According to the running state of the high-speed bearing, the bearing degradation signal x(t) is obtained according to the periodic motion and degradation characteristics of the bearing, and the bandwidth of the signal is smoothed using the Hanning window function to achieve uniform bandwidth mapping.

[0066] For a vibration signal x(t) of length n, an optimization objective function is constructed. The model is used to calculate the modal function u of the original signal. k and the center frequency ω k Perform minimum optimization. The optimization objective function is defined as:

[0067]

[0068] in Represents the partial derivative operation on time t; δ(t) is the unit impulse function, which is used to localize or filter the signal; is a singular kernel function; * represents a convolution operator; Represents the sum of squares of the signal.

[0069] Establish an optimization constraint function to characterize the relationship between the original signal and the modal function of the original signal:

[0070]

[0071] x(t) represents a vibration signal of length n; u k is the mode function of the original signal.

[0072] After defining the optimization objective function and characterizing the relational constraints, the modal function u of the original signal can be k and the center frequency ω k Perform minimum optimization to solve the modal function u k and the center frequency ω k The solution formulas are:

[0073]

[0074] in, represents the estimated value of the kth mode signal in the frequency domain ω at the n+1th iteration; Represents the estimated value of the input signal x(t) in the frequency domain ω; represents a constraint term; α represents the smoothing index.

[0075] (2) High-dimensional sparse characteristic representation of the life of high-speed heavy-duty transmission systems;

[0076] According to the optimal modal signal u obtained in step (1) k (ω),u k (ω) is more representative of the bearing life characteristics in the long-term trend and is more representative of the optimal modal signal uk (ω) is normalized to avoid the gradient explosion problem. Specifically, the normalized optimal modal signal for

[0077]

[0078] in, represents the standard normalized signal, u k (ω) represents the vibration information of the high-speed bearing, max() represents the maximum value operation, and min() represents the minimum value operation.

[0079] The normalized optimal modal signal is further processed by Gram angular field to achieve high-dimensional representation of one-dimensional signal. High-dimensional representation of one-dimensional signal GADF i，j for:

[0080]

[0081] where GADF is the Gram angular field transform, is the row standard normalized signal, is the column standard normalized signal. The signal after Gram angle field mapping has high dimensional characteristics. According to GADF i，j The spectral characteristics of the multidimensional sparse characteristic Γ are characterized by:

[0082]

[0083] Among them, I is the length of the time series after mapping, J is the dimension of the time series after mapping, and GADF i，j It is a high-dimensional representation of a one-dimensional signal.

[0084] (3) Determination of the spillover time of defects in the transmission system of high-speed heavy machinery;

[0085] Based on the original bearing degradation signal x(t), the multi-dimensional sparse characteristics extracted in step (2) are rearranged in time sequence to obtain the time sequence high-dimensional sparse characteristics Γ(t), whose envelope energy E(t) can be expressed as:

[0086]

[0087] Among them, Γ(t) represents the time series high-dimensional sparse characteristics, t represents the time series information of the high-dimensional sparse characteristics, E(t) represents the envelope energy of the high-dimensional sparse characteristics, and T is the current moment information.

[0088] Weibull distribution is suitable for describing the failure behavior of mechanical equipment such as bearings. The probability density function that obeys the Weibull distribution is defined as:

[0089]

[0090] Among them, T is the current time information, β is the shape parameter, and η is the scale parameter. The optimal values ​​of the shape parameter β and the scale parameter η are determined by the linear regression method:

[0091] 1n(-ln(S(T)))=βln(T)-βln(η)

[0092] The constructed probability density function can be used to derive the survival function S(T) that represents the probability of no damage occurring before time T and the risk function H(T) that represents the probability of instantaneous damage occurring at time T:

[0093]

[0094] Where T is the current moment information, β is the shape parameter, and η is the scale parameter.

[0095] Determining the survival probability threshold (usually 95%) can lead to the defect spillover time T FFT :

[0096] T FFT =min{T|S(T)≤S th}

[0097] (4) Establishment of stacked guidance layer network model;

[0098] According to step (3), the high-dimensional sparse feature representation Γ(t) after the defect overflow time T FFT After the determination, in order to deeply mine the deep sparse characteristic information, the guided layer network model is established. The model of the guided layer network is based on the single-layer recursive layer network. First, the core formula of the single-layer recursive layer network is constructed:

[0099] h t =σ(W h h t-1 +W x Γ(t)+b h )

[0100] Where Γ(t) represents the high-dimensional sparse nature of the time series, h t represents the temporal hidden layer, σ-dimensional nonlinear activation function, W h , W x 、b h Represents the weight and bias parameters of the network.

[0101] The guided layer network is constructed by stacking multiple layers of recursive units in time sequence, and the output of each layer of recursive units is used as the input of the next layer, thereby realizing multi-layer feature recognition. Its formula is:

[0102]

[0103] Among them, Wh , W x , b h represents the weight and bias parameters of the network, are the weight and bias parameters of the lth layer.

[0104] For the constructed regression layer network, the regression task loss function is introduced Optimize and adjust parameters:

[0105]

[0106] Among them, Γ(t) represents the high-dimensional sparse characteristics of time series, represents the trait mean, Represents the regression task loss function.

[0107] (5) Life assessment of high-speed heavy machinery transmission system based on stacked guidance layer network.

[0108] According to the guided layer network model established in step (4), the life of the high-speed heavy machine transmission system is evaluated. The input high-dimensional sparse features are linearly transformed through the first fully connected layer of the guided layer network model to obtain the life evaluation result:

[0109] θ=g(h(Γ(t);θ h );θ g )

[0110] Among them, Θ is the life assessment result of the high-speed heavy machine transmission system, g() represents the activation function, h() represents the hidden layer operation, and θ is the interference function.

Claims

1. A method for evaluating the life of a high-speed heavy machine transmission system based on a sparse characteristic-guided layer network, characterized in that: The steps include: (1) The deterioration data disturbance of high-speed heavy-duty machine transmission system is weakened; (2) High-dimensional sparse characteristic representation of the life of high-speed heavy-duty transmission systems; (3) Determination of the spillover time of defects in the transmission system of high-speed heavy machinery; (4) Establishment of stacked guidance layer network model; (5) Life assessment of high-speed heavy machinery transmission system based on stacked guidance layer network.

2. The life assessment method for high-speed heavy machine transmission system based on sparse characteristic-guided layer network according to claim 1 is characterized in that: The step (1) includes first defining the optimization objective function and characterizing the relationship constraint, and then performing the modal function u of the original signal. k and the center frequency ω k Perform minimum optimization solution.

3. The life assessment method for high-speed heavy machine transmission system based on sparse characteristic-guided layer network according to claim 2 is characterized in that: Defining the optimization objective function and characterizing the relational constraints include: According to the running state of the high-speed bearing, the periodic motion and degradation characteristics of the bearing, the bearing degradation signal x(t) is obtained, and the bandwidth of the signal is smoothed using the Hanning window function to achieve uniform bandwidth mapping. For a vibration signal x(t) of length n, an optimization objective function is constructed. The model is used to calculate the modal function u of the original signal. k and the center frequency ω k For minimal optimization, the optimization objective function is defined as: in Represents the partial derivative operation on time t; δ(t) is the unit impulse function, which is used to localize or filter the signal; is a singular kernel function; * represents a convolution operator; represents the sum of squares of the signal; Establish an optimization constraint function to characterize the relationship between the original signal and the modal function of the original signal: x(t) represents a vibration signal of length n; u k is the mode function of the original signal.

4. The method for evaluating the life of a high-speed heavy machine transmission system based on a sparse characteristic-guided layer network according to claim 2 is characterized in that: The modal function u of the original signal k and the center frequency ω k The formulas for minimum optimization solutions are: in, represents the estimated value of the kth mode signal in the frequency domain ω at the n+1th iteration; Represents the estimated value of the input signal x(t) in the frequency domain ω; represents a constraint term; α represents the smoothing index.

5. The method for evaluating the life of a high-speed heavy machine transmission system based on a sparse characteristic-guided layer network according to claim 1 is characterized in that: The step (2) comprises: According to the optimal modal signal u obtained in step (1) k (ω),u k (ω) is more representative of the bearing life characteristics in the long-term trend and is more representative of the optimal modal signal u k (ω) is normalized to avoid the gradient explosion problem. Specifically, the normalized optimal modal signal for: in, represents the standard normalized signal, u k (ω) represents the vibration information of the high-speed bearing, max() represents the maximum value operation, and min() represents the minimum value operation.

6. The method for evaluating the life of a high-speed heavy machine transmission system based on a sparse characteristic-guided layer network according to claim 5 is characterized in that: The normalized optimal modal signal is processed by dimensionality increase through the Gram angle field to achieve a high-dimensional representation of the one-dimensional signal. The high-dimensional representation form of the one-dimensional signal is GADF i，j for: Among them, GADF is the Gram angular field transform, is the row standard normalized signal, is the column standard normalized signal; the signal after Gram angle field mapping is calculated according to GADF i，j The spectral characteristics of the multidimensional sparse characteristic Γ are characterized by: Among them, I is the length of the time series after mapping, J is the dimension of the time series after mapping, and GADF i，j It is a high-dimensional representation of a one-dimensional signal.

7. The method for evaluating the life of a high-speed heavy machine transmission system based on a sparse characteristic-guided layer network according to claim 1 is characterized in that: The step (3) comprises: Based on the original bearing degradation signal x(t), the multi-dimensional sparse characteristics extracted in step (2) are rearranged in time sequence to obtain the time sequence high-dimensional sparse characteristics Γ(t), whose envelope energy E(t) can be expressed as: Among them, Γ(t) represents the time series high-dimensional sparse characteristics, t represents the time series information of the high-dimensional sparse characteristics, E(t) represents the envelope energy of the high-dimensional sparse characteristics, and T is the current moment information; Weibull distribution is suitable for describing the failure behavior of mechanical equipment such as bearings. The probability density function that obeys the Weibull distribution is defined as: Among them, T is the current time information, β is the shape parameter, and η is the scale parameter. The optimal values ​​of the shape parameter β and the scale parameter η are determined by the linear regression method: ln(-ln(S(T)))=βln(T)-βln(η) The constructed probability density function can be used to derive the survival function S(T) that represents the probability of no damage occurring before time T and the risk function H(T) that represents the probability of instantaneous damage occurring at time T: Where T is the current moment information, β is the shape parameter, and η is the scale parameter; Determining the survival probability threshold can lead to the defect spillover time T FFT : T FFT =min{T|S(T)≤S th } Among them, the survival probability threshold is 95%.

8. The method for evaluating the life of a high-speed heavy machine transmission system based on a sparse characteristic-guided layer network according to claim 1 is characterized in that: The step (4) comprises: According to step (3), the high-dimensional sparse feature representation Γ(t) after the defect overflow time T FFT After confirmation, in order to deeply mine the deep sparse characteristic information and establish the guided layer network model, the guided layer network model takes the single-layer recursive layer network as the core, and constructs the core formula of the single-layer recursive layer network: h t =σ(W h h t-1 +W x C(t)+b h ) Where Γ(t) represents the high-dimensional sparse nature of the time series, h t represents the temporal hidden layer, σ-dimensional nonlinear activation function, W h , W x , b h Represents the weight and bias parameters of the network.

9. The method for evaluating the life of a high-speed heavy machine transmission system based on a sparse characteristic-guided layer network according to claim 8 is characterized in that: The establishment of the guidance layer network model includes: The guided layer network is constructed by stacking multiple layers of recursive units in time sequence. The output of each layer of recursive units is used as the input of the next layer, thereby realizing multi-layer feature recognition. The formula is: Among them, W h , W x , b h represents the weight and bias parameters of the network, are the weight and bias parameters of the ,th layer. For the constructed regression layer network, the regression task loss function is introduced Optimize and adjust parameters: Among them, Γ(t) represents the high-dimensional sparse characteristics of time series, represents the trait mean, Represents the regression task loss function.

10. The method for evaluating the life of a high-speed heavy machine transmission system based on a sparse characteristic-guided layer network according to claim 1 is characterized in that: The step (5) comprises: According to the guided layer network model established in step (4), the life of the high-speed heavy machine transmission system is evaluated. The input high-dimensional sparse features are linearly transformed through the first fully connected layer of the guided layer network model to obtain the life evaluation result: Θ=g(h(Γ(t);θ h );i g ) Among them, Θ is the life assessment result of the high-speed heavy machine transmission system, g() represents the activation function, h() represents the hidden layer operation, and θ is the interference function.

Images