Wheel polygon monitoring method based on improved adaptive wave trap

Abstract

The invention relates to the technical field of wheel monitoring, in particular to a wheel polygon monitoring method based on an improved self-adaptive wave trap, which can accurately and efficiently represent and quantify wheel polygon characteristics in real time and realize wheel state monitoring. The method is applied to the state monitoring and detection application of the polygon of the high-speed train wheel with harmonic characteristics for the first time. By introducing an adjustable factor into a traditional NG algorithm, compared with a traditional ANF, the improved MANF has a higher convergence speed and still keeps good robustness in a large-noise environment, wheel polygon features can be accurately and efficiently represented and quantified in real time, and wheel state monitoring is achieved.

Description

Technical Field

[0001] The invention relates to the technical field of wheel monitoring, and in particular to a wheel polygon monitoring method based on an improved adaptive notch filter. Background Art

[0002] Wheels are the main load-bearing components of high-speed trains, bearing almost all the weight of the train and completing the rotation on the rails. With the continuous acceleration of railway transportation, the interaction force between the train and the rails is gradually increasing, which is easy to cause wheel defects. Among them, wheel polygon is a particularly prominent problem, which not only increases the wheel-rail impact force, causes track wear, local damage to the wheel tread and rails, but also causes major safety accidents such as train derailment in severe cases. Under ideal conditions, when wheel polygons exist, the spectral characteristics of its vibration signal are usually expressed as integer multiples of the rotation frequency harmonic form proportional to the order. Therefore, real-time tracking and monitoring of the vibration characteristic information generated by the wheels during the operation of the train can estimate the order and wear amplitude of the wheels at the current stage, and can effectively evaluate and diagnose the wheel status in real time.

[0003] At present, wheel polygon diagnosis methods based on vibration signals mainly include time-frequency analysis methods such as STFT, wavelet transform, EMD, or machine learning methods such as SVM and CNN. Although the above methods can extract features or intelligently detect wheel out-of-roundness or polygonal status to a certain extent, it is difficult to obtain the key features (order and amplitude) of wheel polygons in one operation, and the required computational cost is relatively high, which cannot meet the real online monitoring needs.

[0004] As a time domain filtering method, the adaptive notch filter (ANF) can change the notch frequency with the frequency of the input signal. It can not only be used to process signals with unknown or changing frequencies, but also requires only a few adjustment parameters to achieve good signal detection effects. It has the advantages of simple structure and low computational complexity. It is suitable for wheel polygons, which require real-time extraction and are susceptible to vehicle speed and fluctuations. The extraction and tracking of characteristic frequencies. Compared with other signal analysis methods, ANF does not have the frequency leakage and strict time window requirements of DFT, the basis function selection problem of wavelet transform algorithm, and the amount of calculation is much smaller than that of neural network algorithm. It has become a popular online signal analysis and harmonic detection algorithm in recent years. In view of the fact that ANF has very good instantaneous and frequency following capabilities, the amount of calculation and analysis efficiency meet the needs of online signal monitoring and processing, and it is used for online detection and real-time characteristic frequency estimation of wheel polygon signals with harmonic characteristics.

[0005] In addition, the adaptive update algorithm is the key and core of the ANF influencing the iterative update capability of the filter coefficients. Due to its simple structure, fast convergence speed and good convergence performance, the adaptive notch filter method based on PG (plain gradient algorithm) and NG (normalized gradient algorithm) is frequently used in signal detection applications with high real-time requirements. The NG algorithm uses a low-pass filtered gradient signal based on PG, which improves convergence and accuracy, but its performance is significantly reduced in low signal-to-noise ratio environments, and is slightly inferior in railway wheel monitoring applications with high background noise. Summary of the invention

[0006] In view of this, the present invention provides a wheel polygon monitoring method based on an improved adaptive notch filter, which can accurately, efficiently and in real time characterize and quantify wheel polygon features to achieve wheel status monitoring.

[0007] To achieve the above purpose, the technical solution of the present invention is as follows:

[0008] The wheel polygon monitoring method based on the improved adaptive notch filter is used to extract the characteristics of the highest energy frequency component in the signal in real time, and the characteristic frequency and characteristic amplitude in the wheel polygon vibration signal with time-varying characteristics are calculated. The polygon order is then obtained in combination with the corresponding speed information, and finally the wheel polygon state is evaluated according to the amplitude.

[0009] The specific implementation steps are as follows:

[0010] Step 1: Obtain vibration acceleration data and train running speed;

[0011] Step 2: Processing the vibration acceleration data using an improved adaptive notch filter to estimate the main harmonic frequencies and their corresponding harmonic energies in the signal in real time;

[0012] Step 3: Calculate the wheel rotation frequency based on the harmonic frequency and the train running speed, and determine the order of the wheel polygon;

[0013] Step 4: According to the harmonic energy and the order of the wheel polygon, determine whether the wheel is in a polygonal state and the specific wear condition of the polygon.

[0014] The improved adaptive notch filter MANF optimizes the normalized gradient algorithm by introducing an adjustable factor.

[0015] The harmonic frequency and the corresponding harmonic energy are calculated by the following formula:

[0016] Harmonic frequency f h (n)

[0017]

[0018] Among them, f s The collected vibration acceleration data x(n) is used to estimate the main harmonic frequency estimation coefficient c in real time through MANF. h (n);

[0019] The corresponding harmonic energy P h (n) is:

[0020] where λ∈(0,1] is the smoothing factor.

[0021] Among them, the wheel speed Where v(n) is the velocity signal and D is the known wheel diameter.

[0022] Among them, the harmonic amplitude with the largest energy estimated in the signal is The corresponding order O th (n) represents the frequency value f of the harmonic with the highest energy in the tracking signal h (n) and the instantaneous frequency value, that is,

[0023]

[0024] Among them, through the order O th (n) Harmonic amplitude A h (n) By comparing with the empirical values, it is determined whether the wheel is in a polygonal state and the corresponding polygonal wear condition, so as to realize online monitoring and evaluation of the polygonal state.

[0025] Beneficial effects:

[0026] 1. This invention applies it to the condition monitoring and detection of high-speed train wheel polygons with harmonic characteristics for the first time. By introducing an adjustable factor into the traditional NG algorithm, the improved MANF has a faster convergence speed and maintains good robustness in a high noise environment compared to the traditional ANF. It can accurately, efficiently and in real time characterize and quantify wheel polygon features to achieve wheel condition monitoring.

[0027] 2. In order to improve the applicability of ANF in wheel feature monitoring, the method of the present invention designs and improves the adaptive update algorithm, introduces an adjustable factor in the filter function to enhance its detection capability in a low signal-to-noise ratio environment, and realizes online estimation and tracking of the most prominent characteristic frequencies and frequency amplitudes in the vibration signal of the train during operation.

[0028] 3. The method of the present invention can not only make a preliminary judgment on the existence of wheel polygons, but also combine the rotation frequency information to effectively characterize and distinguish polygon fault information of different orders, different polygon wear degrees and different speeds. Compared with the traditional vehicle-mounted detection method based on STFT, ANF has higher computational efficiency, time-frequency resolution and stronger focusing ability, which provides a new idea for the implementation of wheel polygon monitoring algorithm on vehicle-mounted equipment.

[0029] 4. Based on the characteristic of MANF that can extract the highest energy frequency component in the signal in real time, the method of the present invention calculates the characteristic frequency and characteristic amplitude in the wheel polygon vibration signal with time-varying characteristics, and then obtains the polygon order in combination with the corresponding speed information. Finally, the wheel polygon state is evaluated according to the amplitude. This is a new attempt of adaptive notch filter in the field of high-speed railway. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 The figure is a flow chart of a method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0031] The present invention is described in detail below with reference to the accompanying drawings and embodiments.

[0032] The principle of the method of the present invention is analyzed as follows:

[0033] An adaptive notch filter is a special form of band-stop filter that filters by attenuating the amplitude of the notch frequency in the signal, while retaining the remaining components without distortion. The notch frequency can be automatically adjusted according to the output error result and configured adaptively. The method of the present invention utilizes this idea. The notch frequency matched by the ANF is the most prominent frequency component in the signal. By extracting the frequency and amplitude of this feature, the real-time status of the current wheel polygon can be known and effectively evaluated. An adaptive notch filter usually uses a second-order IIR in the form of zero-pole constraints as the notch filter structure, and its transfer function (z transform) and coefficient update expression are expressed as:

[0034]

[0035] c=-2cosω (2)

[0036] Among them, z represents the variable of z transformation; ρ represents the size of the polar radius, which usually determines the notch width; c represents the update coefficient, from which the instantaneous value of the notch frequency ω can be estimated. When the input signal x(n) enters the notch filter unit, the bandpass signal s(n), the residual signal e(n) and the gradient signal g(n) will be output. The bandpass signal s(n) can be realized by subtracting the input signal from the notch signal. The transfer function of the bandpass filter and the s(n) signal expression can be expressed as:

[0037]

[0038] s(n)=x(n)-e(n) (4)

[0039] Where n represents the discrete time index; the expression of the residual signal e(n) can usually be realized by the difference equation of the notch filter. By using the pseudo-linear approximation, the gradient component can be obtained accordingly.

[0040]

[0041] Accordingly, if NG is used to update the notch coefficient c(n), the expression is as follows:

[0042]

[0043] G(n)=αG(n-1)+(1-α)g 2 (n) (8)

[0044] Among them, μ is the step size parameter that controls the convergence speed; ∈ is a small positive number introduced to avoid division by zero; α∈[0,1) represents a forgetting factor that is less than but very close to 1.

[0045] In order to extend the robustness of NG in low signal-to-noise ratio, an adjustable factor is introduced and used as an update method of modified ANF (MANF) to iteratively update the frequency coefficients. The specific expression formula is as follows:

[0046]

[0047] Among them, γ (≥1) is a variable parameter and a key factor in the update algorithm in MANF. If γ = 1, the improved update algorithm is the conventional NG algorithm. If γ can be selected correctly, the improved NG algorithm can be significantly better than the conventional NG algorithm in a wider signal-to-noise ratio and frequency range.

[0048] Finally, according to the real-time estimation of the main harmonic frequency coefficient c, and c=-2cosω, the corresponding real-time frequency can be obtained.

[0049] When a wheel has a polygonal defect, the vibration signal will be accompanied by characteristic harmonics that are proportional to the order and have prominent energy. Therefore, tracking and identifying the characteristic frequency with the largest energy in the vibration signal during train operation can monitor the polygonal state of the wheel. Based on the characteristic of MANF that can extract the highest energy frequency component in the signal in real time, the characteristic frequency and characteristic amplitude in the wheel polygonal vibration signal with time-varying characteristics are calculated, and then the polygon order is obtained in combination with the corresponding speed information. Finally, the wheel polygonal state is evaluated according to the amplitude. Combined with the above principle analysis, the wheel polygon monitoring method based on the improved adaptive notch filter of the present invention is proposed, and the process is as follows: Figure 1 As shown, the specific implementation steps are as follows:

[0050] Step 1: Data acquisition: The vibration acceleration data is acquired through the vibration signal sensor connected to the wheel axle box, and the running speed signal is acquired from the speed collector.

[0051] Step 2: Harmonic feature extraction and estimation. The collected vibration acceleration data x(n) is used by MANF to estimate the main harmonic frequency estimation coefficient c in real time. h (n), joint coefficient and frequency conversion formula c = -2cosω, real-time estimated harmonic frequency f h (n) can be calculated as:

[0052]

[0053] Among them, f s is the sampling frequency. The harmonic frequency obtained at this time also represents the most obvious characteristic frequency in the signal. According to formula (5), the harmonic signal s after removing irrelevant components can be obtained: h (n) is estimated to be:

[0054] s h (n) = x(n) - e(n) (16)

[0055] The corresponding harmonic energy P h (n) can be calculated as:

[0056]

[0057] where λ∈(0,1] is the smoothing factor.

[0058] In addition, the wheel rotation frequency f can be calculated from the speed signal v(n) and the known wheel diameter D: r (n) is:

[0059]

[0060] Step 3: Based on the instantaneous harmonic energy P estimated in real time in step 2h (n), instantaneous harmonic frequency f h (n) and instantaneous transfer frequency f r (n) The characteristic quantity used to characterize the real-time state of the wheel polygon can be obtained. The harmonic amplitude A with the largest energy estimated in the signal h (n), can reflect the characteristic frequency size information of the wheel polygon, which is expressed as:

[0061]

[0062] The corresponding order O th (n) can be expressed as the frequency value f of the highest energy harmonic in the tracking signal h (n) Ratio to instantaneous frequency value:

[0063]

[0064] Step 4: Finally, through order O th (n) Harmonic amplitude A h (n) By comparing with the empirical values, it is determined whether the wheel is in a polygonal state and the corresponding polygonal wear condition, so as to realize online monitoring and evaluation of the polygonal state.

[0065] In summary, the above are only preferred embodiments of the present invention and are not intended to limit the protection scope of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A wheel polygon monitoring method based on an improved adaptive notch filter, characterized in that: The characteristics of the highest energy frequency component in the signal are extracted in real time, the characteristic frequency and characteristic amplitude in the wheel polygon vibration signal with time-varying characteristics are calculated, and then the polygon order is obtained in combination with the corresponding speed information. Finally, the wheel polygon state is evaluated according to the amplitude.

2. The method according to claim 1, characterized in that The specific implementation steps are as follows: Step 1: Obtain vibration acceleration data and train running speed; Step 2: Processing the vibration acceleration data using an improved adaptive notch filter to estimate the main harmonic frequencies and their corresponding harmonic energies in the signal in real time; Step 3: Calculate the wheel rotation frequency based on the harmonic frequency and the train running speed, and determine the order of the wheel polygon; Step 4: According to the harmonic energy and the order of the wheel polygon, determine whether the wheel is in a polygonal state and the specific wear condition of the polygon.

3. The method according to claim 2, characterized in that The improved adaptive notch filter MANF optimizes the normalized gradient algorithm by introducing an adjustable factor.

4. The method according to any one of claims 1 to 3, characterized in that , the harmonic frequency and the corresponding harmonic energy are calculated by the following formula: Harmonic frequency f h (n) Among them, f s The sampling frequency is the collected vibration acceleration data x(n), and the main harmonic frequency estimation coefficient c in the signal is estimated in real time by MANF. h (n); The corresponding harmonic energy P h (n) is: where λ∈(0,1] is the smoothing factor.

5. The method according to claim 4, characterized in that Wheel speed Where v(n) is the velocity signal and D is the known wheel diameter.

6. The method according to claim 4 or 5, characterized in that , the estimated maximum harmonic amplitude in the signal The corresponding order O th (n) represents the frequency value f of the harmonic with the highest energy in the tracking signal h (n) and the instantaneous frequency value, that is, 7. The method according to claim 6, characterized in that , through order O th (n) Harmonic amplitude A h (n) By comparing with the empirical values, it is determined whether the wheel is in a polygonal state and the corresponding polygonal wear condition, so as to realize online monitoring and evaluation of the polygonal state.

Images