Fault diagnosis method for belt conveyor

Abstract

The invention relates to a belt conveyor fault diagnosis method based on dynamic window adjustment and multi-scale analysis. A traditional method adopts a fixed window to analyze a current signal, and has time-frequency contradictions: a long window is slow in response although the frequency resolution is improved, and a short window is fast in response but low in spectrum resolution and strong in noise interference. According to the method, dynamic window adjustment and multi-scale segmentation are introduced, the window length is adaptively selected according to local statistical characteristics of signals, analysis is performed under different scales, and amplitude and phase information of the modulation signals is extracted at the same time through multi-scale discrete Fourier transform and sideband square modulation double-spectrum calculation. The feature stability is further improved through set averaging and weighted data fusion, and parameters are dynamically updated by using an online self-calibration mechanism to adapt to non-stationary working conditions. The method solves the time-frequency contradiction of a traditional method, achieves the real-time and accurate diagnosis of the faults of the belt conveyor, and has a remarkable engineering application value.

Description

Technical Field

[0001] The present invention relates to a fault diagnosis method for a belt conveyor, belonging to the technical field of fault diagnosis. Background Art

[0002] Most of the existing fault diagnosis methods for belt conveyors adopt non-invasive current signal acquisition, and extract modulation sideband features based on the discrete Fourier transform (DFT) and the bispectrum algorithm to monitor the states of belt tension, slack, misalignment of transmission, etc. However, when the traditional method realizes high frequency resolution, a relatively long window is usually adopted to improve the signal-to-noise ratio and spectral accuracy. However, the long window leads to a decrease in real-time response ability; on the contrary, although a shorter window can capture transient fault information, it is prone to spectral leakage and noise interference, resulting in unclear fault features. This contradiction between "long window high resolution" and "short window real-time response" urgently needs to find a suitable countermeasure. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to design a fault diagnosis method for a belt conveyor to overcome the deficiencies of the prior art.

[0004] The technical solution of the present invention is as follows: A fault diagnosis method for a belt conveyor is provided, and the method includes the following steps:

[0005] S1. Collect the current signal of the driving motor of the belt conveyor;

[0006] S2. Perform local spectral analysis on the collected current signal within a time window, use a window function to weight the signal within the window, and obtain a local spectral representation. At the same time, calculate the energy and fluctuation index within the window to judge the stationarity of the signal during this period;

[0007] S3. Adaptively determine the length of the window according to the changes of the local energy and fluctuation index, and segment the original signal at multiple scales;

[0008] S4. Apply the corresponding window function to the signal segments at each scale in step 3, then perform discrete Fourier transform to obtain the frequency domain representation at each scale, and extract the amplitude and phase information of each frequency component respectively;

[0009] S5. At each scale, for the signal at the key frequency, extract the upper sideband and lower sideband components, and construct sideband square modulation bispectrum data by using the sideband amplitude and phase information;

[0010] S6. Perform weighted average fusion on the bispectrum data obtained from each scale and each data segment by using a preset weight to form an overall fused bispectrum representation, and extract the amplitude feature and phase feature reflecting the belt state therefrom;

[0011] S7. Compare the amplitude characteristics and phase characteristics with a pre-established standard model to determine whether there are faults such as belt tension, slack, or incorrect transmission alignment, and feedback the fault determination result to the monitoring system in real time.

[0012] Further, the window lengths of multiple scales in step S3 are evenly distributed between the shortest window and the longest window.

[0013] Further, the window function used in step S2 is a Hanning window function.

[0014] Further, a fixed percentage overlap technique is adopted during signal segmentation, and the overlap rate is approximately 50%.

[0015] Further, the method for determining the stationarity of the signal in step S2 is as follows: By calculating the ratio of the energy of the local signal to the fluctuation index, when the ratio exceeds a preset threshold, the signal is determined to be an unsteady region, and a short window is adopted in this region, otherwise a long window is adopted.

[0016] Further, the sideband square modulation bispectrum data constructed in step 5 is obtained by multiplying the upper sideband amplitude by the lower sideband amplitude and combining the sum of the phases of both sidebands minus twice the carrier phase. The bispectrum data can distinguish the signal characteristics of pure amplitude modulation, pure phase modulation, and the mixed modulation of both.

[0017] Further, the weighted average fusion adopted in step 6, where the weights of each scale are determined according to the signal-to-noise ratio of each scale signal segment.

[0018] Further, the fault determination of the belt state in step 7 is performed by comparing the amplitude characteristics and phase characteristics at the key frequencies after fusion with the reference values obtained through pre-calibration.

[0019] The beneficial effects of the present invention are: Compared with the prior art,

[0020] 1) By local statistical analysis, the present invention dynamically adjusts the window length, so that a longer window is adopted in the steady signal region to obtain high spectral resolution, while in the fault-sensitive region, it quickly switches to a short window to capture transient information. This adaptive window adjustment strategy effectively solves the inherent contradiction of the fixed window in time-frequency analysis in the traditional method, not only ensures the accurate capture of signal details, but also realizes fast response, thereby improving the accuracy and timeliness of fault diagnosis;

[0021] 2) The present invention fuses data of different scales by using a weighted average method. In this way, not only the interference of random noise on fault feature extraction is effectively suppressed, but also the stability and robustness of fault features are significantly improved. The multi-scale data fusion takes into account the characteristic information of signals at different scales, making the final fusion result more comprehensive and accurate, being able to more reliably reflect the actual operating state of the belt conveyor, and reducing the possibility of misjudgment and missed judgment;

[0022] 3) The present invention can simultaneously extract the amplitude and phase information of the modulation signal through sideband square modulation bispectrum calculation. This is of great significance for the accurate diagnosis of belt faults. By analyzing the changes in amplitude and phase, the fault type and degree of the belt can be accurately judged, providing a strong basis for taking timely maintenance measures. This bispectrum modulation analysis method makes up for the deficiencies of traditional single amplitude or phase analysis, improving the accuracy and reliability of fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 is a flowchart of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0024] To make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail with reference to the drawings of this specification.

[0025] Refer to Figure 1 , this embodiment provides a belt conveyor fault diagnosis method, including the following steps: Step S1: Signal acquisition

[0026] Connect a current clamp or other non-invasive sensor to the driving motor of the belt conveyor, and collect the current signal of the driving motor of the belt conveyor in real time, denoted as x(t), t ∈ [0, T total , where: x(t) is the current signal; t is the time variable; T total is the total acquisition duration, and then use a high-speed data acquisition instrument (the sampling frequency can be selected as 50 kHz or higher) to transmit the collected signal to the signal processing computer.

[0027] Step S2: Preliminary short-time local analysis

[0028] Select a short-time window (for example, use a shorter sampling segment) from the continuously collected current signals for preliminary analysis, use methods such as short-time Fourier transform (STFT) or wavelet transform to detect transient changes or local anomalies, and judge whether there is a fault-sensitive area in the current signal segment according to the short-time analysis results. By calculating the ratio of the energy of the local signal to the fluctuation index, when this ratio exceeds a preset threshold, judge that the signal is an unstable area, and thus use a short window in this area, otherwise use a long window, providing a basis for subsequent window length adjustment. Select a short-time window Δt sThe signal segment within (e.g., 20 to 50 milliseconds),

[0029] x s x(t) = x(t), t ∈ [t0, t0 + Δt s ,

[0030] Perform a short-time Fourier transform (STFT) on it:

[0031]

[0032] where: w s (t) is a short-time window function (e.g., Hanning window); f is the frequency variable.

[0033] Simultaneously calculate the local energy E(t0) and variance σ 2 (t0):

[0034]

[0035] where μ(t0) is the mean value of the signal within the window. Use E(t0) and σ 2 (t0) to judge local stationarity: If the energy fluctuation or variance is large, it is considered that there is a transient or local fault-sensitive region.

[0036] Step S3: Dynamic window adjustment and multi-scale segmentation

[0037] According to the local signal characteristics obtained in step S2, adopt a dynamic window adjustment algorithm to determine the adaptive window length:

[0038] · For the signal stationary region, preferably use a long window to obtain high frequency resolution;

[0039] · For the mutation or transient event region, use a shorter window to improve the time-domain response speed.

[0040] Meanwhile, to balance the global diagnostic accuracy and local fault detection, implement a multi-scale segmentation strategy:

[0041] · For the same original signal, divide windows of multiple scales (e.g., short windows and long windows) simultaneously according to a predetermined rule, and independently calculate the spectral information at each scale. The window lengths of multiple scales are evenly distributed between the shortest window and the longest window;

[0042] · The multi-scale information is fused through subsequent processing to ensure both sensitive detection of local details and the stability of the overall spectrum.

[0043] According to the statistical indicators in step S2, adopt the following adaptive window strategy:

[0044] · Define a short window T S and a long window T LThe preset value (e.g., T S = 20 ms, T L = 100 ms).

[0045] · Let the local non-stationarity index γ(t0) = σ 2 (t0) / E(t0);

[0046] · Define the threshold γ th , when γ(t0) > γ th , use the short window T S , otherwise use the long window T L ;

[0047] · To obtain richer time-frequency information, set multiple scales k = 1, 2, …, K on the same signal segment, and let

[0048]

[0049] · Thus, a multi-scale window set is generated.

[0050] Segment the original signal x(t), and denote the i-th data segment at scale k as

[0051] x i,k (t), t ∈ [t i , t i + T k ,

[0052] where t i is determined according to the window overlap rate η (such as 50% overlap).

[0053] Step S4: Multi-scale discrete Fourier transform

[0054] For each dynamically adjusted data segment obtained in S3, apply an appropriate window function (such as a Hanning window) respectively, and perform a discrete Fourier transform (DFT) on each scale data segment in turn to obtain the frequency-domain signal X(f); during this process, record the spectral resolution and amplitude information at different window lengths simultaneously to provide a data basis for subsequent feature extraction.

[0055] Apply the corresponding window function w i,k (t) to each data segment x k (t) and then perform DFT:

[0056]

[0057] Decompose the amplitude spectrum and phase spectrum:

[0058] A i,k (f) = |X i,k (f)|, φ i,k (f) = argXi,k (f).

[0059] Step S5: Sideband Square Modulation Bispectrum Calculation

[0060] At each scale, for the signals at the key frequencies, extract the upper sideband and lower sideband components, and construct the sideband square modulation bispectrum data using the sideband amplitude and phase information.

[0061] Set the modulation sideband frequency offset Δf, which is related to the motor speed and the belt transmission structure. For each data segment, extract the upper sideband and lower sideband near the key frequency f:

[0062] X i,k (f + Δf), X i,k (f - Δf).

[0063] The sideband square modulation bispectrum data (sideband square modulation bispectrum composite quantity) is calculated by multiplying the upper sideband amplitude by the lower sideband amplitude and combining the sum of the phases of the two sidebands minus twice the carrier phase. The bispectrum data can distinguish the signal characteristics of pure amplitude modulation, pure phase modulation, and the mixed modulation of both. Specifically, the sideband square modulation bispectrum composite quantity B i,k (f, Δf) is calculated as follows:

[0064] B i,k (f, Δf) = A i,k (f + Δf) · A i,k (f - Δf) · exp{j[φ i,k (f + Δf) + φ i,k (f - Δf) - 2φ i,k (f)]}.

[0065] Where:

[0066] · For pure AM signals, it is expected that φ i,k (f + Δf) + φ i,k (f - Δf) - 2φ i,k (f) ≈ 0;

[0067] · For pure PM signals, it is expected that this value is close to ±π;

[0068] · For mixed modulation, the value falls within the interval [0, ±π].

[0069] Step S6: Ensemble Averaging and Multi-Scale Data Fusion

[0070] The double - spectral data obtained for each scale and each data segment are weighted - averaged and fused using a preset weight to form an overall fused double - spectral representation, from which amplitude features and phase features reflecting the belt state are extracted. The fault determination of the belt state is carried out by comparing the amplitude features and phase features at the key frequencies after fusion with the reference values obtained through pre - calibration.

[0071] Perform weighted - average processing on the multi - scale double - spectral results obtained from S5 to further reduce the interference of random noise; identify the key frequency points with the largest amplitude and closely related to the belt state in the fused spectrum, and extract the corresponding amplitude and phase features; use the pre - established standard model of the belt state to determine the corresponding relationship between the amplitude change and phase drift and the belt fault.

[0072] To suppress random noise, weighted - average the double - spectral data calculated for all scales k and each data segment i. Introduce a scale - weighted factor ω k (which can be determined according to the signal - to - noise ratio of each scale) to obtain the fused double - spectrum:

[0073]

[0074] where N k is the number of data segments at scale k. Extract the amplitude and phase features after fusion:

[0075]

[0076] Step S7: Fault determination and real - time response

[0077] The fault determination of the belt state is carried out by comparing the amplitude features and phase features at the key frequencies after fusion with the reference values obtained through pre - calibration. According to the key features extracted in S6, perform real - time fault - state judgment: when it is detected that the amplitude at the key frequency point decreases, it indicates an increase in belt tension; when the amplitude increases, it indicates a risk of belt slack; when the phase has an abnormal offset and under the condition that the processing parameters remain unchanged, it indicates the existence of dynamic eccentricity or misalignment in the transmission system. Feed the fault diagnosis results back to the monitoring system in real - time and trigger an alarm or an automatic adjustment control strategy (such as adjusting the belt tensioning mechanism) when necessary to ensure the efficient operation of the system.

[0078] Select a characteristic frequency f0 closely related to the belt state (for example, the belt - passing frequency or the frequency calculated from the motor speed and belt geometric parameters). Extract the characteristic amplitude and phase at this frequency:

[0079] A B (f0,Δf),φ B (f0,Δf).

[0080] Based on a pre-established standard model and threshold judgment:

[0081] · When A B (f0, Δf) is significantly lower than the reference value, it indicates an increase in belt tension;

[0082] · When A B (f0, Δf) is significantly higher than the reference value, it indicates belt slack;

[0083] · When φ B (f0, Δf) shows an abnormal deviation under the condition that the processing parameters remain unchanged, it indicates the existence of dynamic eccentricity or transmission misalignment problems.

[0084] The judgment result is fed back to the monitoring system in real time, and the automatic adjustment device (such as adjusting the belt tensioning mechanism) can be triggered to reduce risks.

[0085] Step S8: Online self-calibration and dynamic update

[0086] To adapt to the non-stationarity of the belt conveyor under different working conditions, an online self-calibration mechanism is established: using real-time monitoring data to continuously update the window adjustment parameters and feature thresholds of multi-scale analysis; adopting a feedback control algorithm to dynamically correct the window length and weighting strategy to ensure a high fault detection sensitivity and response speed during the long-term operation of the system. To adapt to the system non-stationarity, an online self-calibration mechanism is established. Let the set of real-time statistical indicators be Θ(t), including local energy, variance, signal-to-noise ratio, etc.:

[0087] Θ(t) = {E(t), σ 2 (t), SNR(t),...}.

[0088] A feedback control algorithm is used to update the window parameters and weighting factors. The specific update formulas are as follows:

[0089]

[0090] Where: and are the old and new window lengths respectively; and are the old and new weighting factors; ΔΘ(t) and ΔSNR(t) are the change amounts of statistical indicators; α and β are adjustment coefficients, determined through experiments to ensure the stability and reliability of the self-calibration process.

[0091] Where the present invention is not described in detail, it is the well-known technology of those skilled in the art. Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and all of them should be covered by the scope of the claims of the present invention.

Claims

1. A fault diagnosis method for a belt conveyor, characterized in that, The method includes the following steps: S1. Collect the current signal of the driving motor of the belt conveyor; S2. Perform local spectral analysis on the collected current signal within a time window, use a window function to weight the signal within the window, and obtain a local spectral representation. At the same time, calculate the energy and fluctuation index within the window to judge the stationarity of the signal during this period; S3. Adaptively determine the length of the window according to the changes of the local energy and fluctuation index, and segment the original signal at multiple scales; S4. Apply the corresponding window function to the signal segments at each scale in step S3 and then perform discrete Fourier transform to obtain the frequency-domain representation at each scale, and extract the amplitude and phase information of each frequency component respectively; S5. At each scale, for the signal at the key frequency, extract the upper sideband and lower sideband components, and use the sideband amplitude and phase information to construct the sideband square modulation bispectrum data; S6. Perform weighted average fusion on the bispectrum data obtained for each scale and each data segment using a preset weight to form an overall fused bispectrum representation, and extract the amplitude characteristics and phase characteristics reflecting the belt state therefrom; S7. Compare the amplitude characteristics and phase characteristics with a pre-established standard model to judge whether the belt has faults such as tension, slack, or misalignment in transmission, and feed back the fault determination result to the monitoring system in real time.

2. The belt conveyor fault diagnosis method according to claim 1, characterized in that, In step S3, the window lengths of multiple scales are evenly distributed between the shortest window and the longest window.

3. The belt conveyor fault diagnosis method according to claim 1, characterized in that, The window function used in step S4 is the Hanning window function.

4. The belt conveyor fault diagnosis method according to claim 1, characterized in that The method for judging the stationarity of the signal in step S2 is: by calculating the ratio of the energy of the local signal to the fluctuation index, when the ratio exceeds a preset threshold, judge that the signal is in an unsteady region, and thus use a short window within this region, otherwise use a long window.

5. The belt conveyor fault diagnosis method according to claim 1, characterized in that In step S5, the sideband square modulation bispectrum data is calculated by multiplying the upper sideband amplitude by the lower sideband amplitude and combining the sum of the phases of the two sidebands minus twice the carrier phase. The bispectrum data can distinguish the signal characteristics of pure amplitude modulation, pure phase modulation, and the mixed modulation of both.

6. The belt conveyor fault diagnosis method according to claim 1, wherein, In the weighted average fusion adopted in step S6, the weights of each scale are determined according to the signal-to-noise ratio of the signal segments at each scale.

7. The belt conveyor fault diagnosis method according to claim 1, characterized in that In step S7, the fault determination of the belt state is made by comparing the amplitude characteristics and phase characteristics at the key frequency after fusion with the reference values obtained by pre-calibration.

Images