Noise feature-based parking equipment mechanical failure prediction method and system

Abstract

The present application relates to a kind of noise feature-based parking equipment mechanical fault prediction method and system, method includes: the real-time mechanical noise signal of parking equipment is collected;Real-time mechanical noise signal is compressed sensing processing, reconstructs to obtain no-fault mechanical noise signal;Real-time mechanical noise signal and no-fault mechanical noise signal are both carried out double feature extraction of resonance peak frequency characteristic parameter and mel frequency cepstrum coefficient;The sum of real-time resonance peak frequency characteristic parameter and real-time mel frequency cepstrum coefficient is subtracted from the sum of no-fault resonance peak frequency characteristic parameter and no-fault mel frequency cepstrum coefficient;Difference and set threshold are compared to carry out fault prediction.Compared with prior art, the present application adopts compressed sensing processing method to reduce the fault characteristic information of real-time mechanical noise signal, by comparing the characteristic parameters of reduced fault characteristic information and the characteristic parameters of original signal, the characteristics of mechanical fault are obtained, realize the automatic prediction of parking equipment mechanical fault.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of equipment detection, in particular to a parking equipment mechanical fault prediction method and system based on noise characteristics. BACKGROUND

[0002] Compared with the traditional garage, the mechanical parking equipment can park more cars in limited space, improve the utilization rate of land resources, and also relieve the pressure between urban planning and environmental development. Many cities are actively applying mechanical multi-layer parking equipment. The mechanical parking equipment manufacturing technology has developed rapidly in China for more than ten years. The vigorous promotion of mechanical parking equipment brings convenience to people, but the faults and accidents also attract great attention from the society. Although most of the faults and accidents are caused by human operation and poor equipment management, but a part of the faults and accidents are caused by product design and quality problems. The diagnosis and prediction of mechanical faults of mechanical parking equipment can effectively reduce the occurrence of mechanical faults and reduce various losses caused by mechanical faults. At present, the prediction of mechanical faults of parking equipment can adopt artificial prediction. For an experienced mechanical engineer, the position and degree of fault can be roughly determined by carefully listening to the mechanical noise of the equipment, but the reliability and accuracy of artificial prediction are low. SUMMARY

[0003] The purpose of the present application is to overcome the defects of the prior art and provide a parking equipment mechanical fault prediction method and system based on noise characteristics.

[0004] The purpose of the present application can be achieved by the following technical solutions:

[0005] A parking equipment mechanical fault prediction method based on noise characteristics, comprising the following steps:

[0006] S1, collecting real-time mechanical noise signals of the parking equipment;

[0007] S2, performing compressed sensing processing on the real-time mechanical noise signals to reconstruct the fault-free mechanical noise signals;

[0008] S3, performing double feature extraction of the real-time mechanical noise signals on the resonance peak frequency characteristic parameters and the mel frequency cepstrum coefficient to obtain the real-time resonance peak frequency characteristic parameters and the real-time mel frequency cepstrum coefficient; at the same time, performing double feature extraction of the fault-free mechanical noise signals on the resonance peak frequency characteristic parameters and the mel frequency cepstrum coefficient to obtain the fault-free resonance peak frequency characteristic parameters and the fault-free mel frequency cepstrum coefficient;

[0009] S4, subtracting the sum of the real-time formant frequency feature parameter and the real-time mel-frequency cepstrum coefficient from the sum of the fault-free formant frequency feature parameter and the fault-free mel-frequency cepstrum coefficient to obtain a difference value;

[0010] S5, comparing the difference value with a set threshold to predict the fault.

[0011] Further, in step S2, the compressive sensing processing comprises the following steps:

[0012] S21, performing frame division and endpoint detection on the real-time mechanical noise signal;

[0013] S22, performing discrete cosine transform on the real-time mechanical noise signal of each frame to obtain DCT domain sparse coefficients and a dynamic threshold, and selecting a set number of multiple DCT domain sparse coefficients to combine the dynamic threshold to perform inverse discrete cosine transform to obtain a sparse pre-processed time domain signal;

[0014] S24, performing compressive projection on the sparse pre-processed time domain signal under a partial random Hadamard observation matrix Φ to obtain an observation vector;

[0015] S25, reconstructing the signal frame with the observation vector under a sparse transform basis matrix Ψ to obtain a fault-free mechanical noise signal.

[0016] Further, in step S21, the endpoint detection adopts a double-threshold endpoint detection method.

[0017] Further, in step S3, the feature extraction of the formant frequency feature parameter comprises the following steps:

[0018] A1, performing frame division and windowing on the real-time mechanical noise signal or the fault-free mechanical noise signal, and then performing fast Fourier transform to obtain a short-time spectrum of the noise signal;

[0019] A2, taking the logarithm of the short-time spectrum of the noise signal and then performing inverse discrete Fourier transform to obtain a cepstrum of the noise signal;

[0020] A3, obtaining the cepstrum of the low-frequency part to perform fast Fourier transform to obtain a logarithmic modulus function, and the peak value of the logarithmic spectrum of the logarithmic modulus function is the formant frequency feature parameter.

[0021] Further, in step S3, the feature extraction of the mel-frequency cepstrum coefficient comprises the following steps:

[0022] B1, performing sampling quantization and endpoint detection on the real-time mechanical noise signal or the fault-free mechanical noise signal to obtain a time domain signal;

[0023] B2, converting the time domain signal into a frequency domain signal through discrete Fourier transform or fast Fourier transform;

[0024] B3, convert the frequency domain signal into a mel spectrum through a mel frequency filter bank;

[0025] B4, log energy processing is performed on the mel spectrum to obtain a log spectrum;

[0026] B5, the log spectrum is converted into a mel frequency cepstral coefficient through a discrete cosine transform.

[0027] Further, in step S4, the real-time formant frequency feature is multiplied by a first weighting coefficient, and the real-time mel frequency cepstral coefficient is multiplied by a second weighting coefficient, and then summed.

[0028] Further, the real-time mechanical noise signal is collected through a high-sensitivity pickup.

[0029] A noise feature-based parking equipment mechanical fault prediction system, comprising a processor and a memory, characterized in that the processor calls a program stored in the memory to realize the noise feature-based parking equipment mechanical fault prediction method as any one of the above.

[0030] Compared with the prior art, the present application has the following beneficial effects:

[0031] 1. The present application adopts a compressive sensing processing method to reduce the fault feature information of the real-time mechanical noise signal, and by comparing the feature parameters of the reduced fault feature information with the feature parameters of the original signal, the feature parameters of the mechanical fault are obtained, thereby realizing the automatic prediction of the mechanical fault of the parking equipment.

[0032] 2. The present application adopts double feature extraction of formant frequency feature parameters and mel frequency cepstral coefficients to improve the redundancy of the feature information, and ultimately improve the accuracy of the prediction diagnosis. BRIEF DESCRIPTION OF DRAWINGS

[0033] Figure 1 The present application is a whole flowchart.

[0034] Figure 2 The present application is a whole flowchart of the compressive sensing algorithm.

[0035] Figure 3 The present application is a flowchart of extracting formant frequency feature parameters.

[0036] Figure 4 The present application is a flowchart of extracting formant frequency feature parameters.

[0037] Figure 5 The present application is a flowchart of extracting mel frequency cepstral coefficient features.

[0038] Figure 6Schematic diagram of the extraction results of Mel-frequency cepstral coefficients.

[0039] Figure 7 This is the fault prediction decision block diagram. DETAILED DESCRIPTION

[0040] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0041] like Figure 1 As shown, this embodiment provides a method for predicting mechanical failures of parking equipment based on noise characteristics, comprising the following steps:

[0042] Step S1: collecting real-time mechanical noise signals of parking equipment. The real-time mechanical noise signals are collected by a high-sensitivity microphone.

[0043] Step S2: Perform compressed sensing processing on the real-time mechanical noise signal to reconstruct a fault-free mechanical noise signal.

[0044] Step S3, performing dual feature extraction of the resonance peak frequency characteristic parameters and the Mel-frequency cepstral coefficients on the real-time mechanical noise signal to obtain the real-time resonance peak frequency characteristic parameters and the real-time Mel-frequency cepstral coefficients; at the same time, performing dual feature extraction of the resonance peak frequency characteristic parameters and the Mel-frequency cepstral coefficients on the fault-free mechanical noise signal to obtain the fault-free resonance peak frequency characteristic parameters and the fault-free Mel-frequency cepstral coefficients.

[0045] Step S4: subtract the sum of the real-time formant frequency characteristic parameters and the real-time Mel-frequency cepstral coefficients from the sum of the fault-free formant frequency characteristic parameters and the fault-free Mel-frequency cepstral coefficients to obtain a difference.

[0046] Step S5: Compare the difference with the set threshold to predict the fault.

[0047] like Figure 2 As shown, the compressed sensing processing of step S2 is expanded as follows: the real-time mechanical noise signal is divided into frames, and pre-processed by segment detection and other methods using the double-threshold endpoint detection method; the sparsity of each frame of the mechanical noise signal in the DCT (discrete cosine transform) domain is then calculated, and then the IDCT (inverse discrete cosine transform) is used to inversely transform it back to the time domain; the pre-processed time domain speech signal frame is compressed and projected under the partially randomized Hadamard measurement matrix Φ to obtain the observation vector; the observation vector is used to initialize the OMP (compressed sensing) reconstruction algorithm: the speech signal frame is reconstructed from the resulting sparse vector, and finally the mechanical noise is reconstructed with some fault characteristics eliminated. The specific steps are as follows:

[0048] Step S21, frame the real-time mechanical noise signal and perform endpoint detection, and the endpoint detection is performed by using a double-threshold endpoint detection method to segment the signal.

[0049] Step S22, perform discrete cosine transform on the real-time mechanical noise signal of each frame to obtain DCT domain sparse coefficients and a dynamic threshold, and perform inverse discrete cosine transform on a plurality of DCT domain sparse coefficients of a set number (the first 300) in combination with the dynamic threshold to obtain a sparse preprocessed time domain signal.

[0050] Step S24, perform compression projection on the sparse preprocessed time domain signal under a partial random Hadamard observation matrix Φ to obtain an observation vector.

[0051] Step S25, reconstruct the signal frame by using the observation vector under a sparse transform basis matrix Ψ to obtain a fault-free mechanical noise signal.

[0052] As shown in Figure 3 and Figure 4 , the extraction of the resonance peak frequency feature parameter in step S3 is expanded as follows: the real-time mechanical noise signal or the fault-free mechanical noise signal is framed, windowed, and subjected to fast Fourier transform (FFT) to obtain a short-time spectrum of the mechanical noise signal. The short-time spectrum is logarithmized and then subjected to inverse discrete Fourier transform (IFFT) to obtain a cepstrum of the mechanical noise signal. The cepstrum separates the spectral envelope of the fundamental harmonic and the vocal tract, and the low-time portion of the cepstrum can analyze the glottal, vocal tract, and radiation information, while the high-frequency portion can analyze the excitation source information. Therefore, the low-time window selects the cepstrum, and the output after FFT is a smoothed logarithmic modulus function. At this time, the smoothed logarithmic spectrum shows the resonance structure of a specific input signal, that is, the peak values of the spectrum substantially correspond to the resonance peak frequency. The peak values in the smoothed logarithmic spectrum are located to obtain the resonance peak. The specific steps are as follows:

[0053] Step A1, frame and window the real-time mechanical noise signal or the fault-free mechanical noise signal, and then perform fast Fourier transform to obtain a short-time spectrum of the noise signal.

[0054] Step A2, logarithmize the short-time spectrum of the noise signal and then perform inverse discrete Fourier transform to obtain a cepstrum of the noise signal.

[0055] Step A3, obtain the cepstrum of the low-frequency portion and perform fast Fourier transform to obtain a logarithmic modulus function. The peak values of the logarithmic spectrum of the logarithmic modulus function are the resonance peak frequency feature parameters.

[0056] As shown in Figure 5As shown, the expansion of the features of the Mel-frequency cepstral coefficients extracted in step S3 is as follows: Mel-frequency cepstral coefficients are characteristic parameters proposed based on the human auditory perception mechanism, which can reflect the voiceprint characteristics of the sound, and it has no premise assumptions and has good recognition performance. The time domain signal of the real-time mechanical noise signal or the fault-free mechanical noise signal is preprocessed by sampling quantization, pre-emphasis framing and windowing, endpoint detection, etc., and is converted into a frequency domain signal through discrete Fourier transform (DFT) or fast Fourier transform (FFT), and then converted into a Mel spectrum through a Mel-frequency filter bank, as shown in FIG. Figure 6 As shown. Then the Mel spectrum is processed by logarithmic energy to obtain the logarithmic spectrum, and finally the Mel frequency cepstral coefficients are obtained by discrete cosine transform (DCT). The specific steps are as follows:

[0057] Step B1: sampling, quantizing, and endpoint detecting the real-time mechanical noise signal or the fault-free mechanical noise signal to obtain a time domain signal.

[0058] Step B2: Convert the time domain signal into a frequency domain signal through discrete Fourier transform or fast Fourier transform.

[0059] Step B3: Convert the frequency domain signal into a Mel spectrum through a Mel frequency filter bank.

[0060] Step B4: perform logarithmic energy processing on the Mel spectrum to obtain a logarithmic spectrum.

[0061] Step B5: Perform discrete cosine transform on the logarithmic spectrum to obtain Mel-frequency cepstral coefficients.

[0062] like Figure 7 As shown, steps S4 and S5 are expanded as follows: The real-time resonance peak frequency characteristics are multiplied by the first weighting coefficient K1, and the real-time Mel-frequency cepstral coefficients are multiplied by the second weighting coefficient K2, and the sum is calculated. The fault-free resonance peak frequency characteristics are multiplied by the first weighting coefficient K1, and the fault-free Mel-frequency cepstral coefficients are multiplied by the second weighting coefficient K2, and the sum is calculated. The calculated result of the fault-free mechanical noise signal branch is subtracted from the calculated result of the real-time mechanical noise signal branch to form a mechanical fault reference measure. A larger value indicates a greater probability of fault. Mechanical faults are diagnosed and issued through threshold comparison.

[0063] The above describes in detail the preferred embodiments of the present invention. It should be understood that those skilled in the art can make numerous modifications and variations based on the concepts of the present invention without inventive effort. Therefore, any technical solutions that can be derived by those skilled in the art through logical analysis, reasoning, or limited experimentation based on the concepts of the present invention and the prior art should be within the scope of protection defined by the claims.

Claims

1. A method for predicting mechanical failure of parking equipment based on noise characteristics, characterized in that: The following steps are involved: S1. Collect real-time mechanical noise signals of parking equipment; S2. Performing compressed sensing processing on the real-time mechanical noise signal to reconstruct a fault-free mechanical noise signal; S3. Perform dual feature extraction of the resonance peak frequency characteristic parameters and the Mel-frequency cepstral coefficients on the real-time mechanical noise signal to obtain the real-time resonance peak frequency characteristic parameters and the real-time Mel-frequency cepstral coefficients; simultaneously, perform dual feature extraction of the resonance peak frequency characteristic parameters and the Mel-frequency cepstral coefficients on the fault-free mechanical noise signal to obtain the fault-free resonance peak frequency characteristic parameters and the fault-free Mel-frequency cepstral coefficients; S4, subtracting the sum of the real-time formant frequency characteristic parameters and the real-time Mel-frequency cepstral coefficients from the sum of the fault-free formant frequency characteristic parameters and the fault-free Mel-frequency cepstral coefficients to obtain a difference; S5. Compare the difference and the set threshold to predict the fault; In step S2, the compressed sensing process includes the following steps: S21, performing framing and endpoint detection on the real-time mechanical noise signal; S22, performing a scattered cosine transform on the real-time mechanical noise signal of each frame to obtain a DCT domain sparse coefficient and a dynamic threshold, selecting a set number of DCT domain sparse coefficients and combining them with the dynamic threshold to perform an inverse scattered cosine transform to obtain a time domain signal after sparsity preprocessing; S24, compressing and projecting the time domain signal after sparsity preprocessing under a partially random Hadamard measurement matrix Φ to obtain an observation vector; S25, reconstructing the signal frame using the observation vector under the sparse transformation basis matrix Ψ to obtain a fault-free mechanical noise signal; In step S4, the real-time resonance peak frequency feature is multiplied by the first weighting coefficient, and the real-time Mel-frequency cepstral coefficient is multiplied by the second weighting coefficient, and then the sum is taken; the fault-free resonance peak frequency feature is multiplied by the first weighting coefficient, and the fault-free Mel-frequency cepstral coefficient is multiplied by the second weighting coefficient, and then the sum is taken.

2. A parking equipment mechanical failure prediction method based on noise characteristics according to claim 1, characterized in that: In step S21, endpoint detection adopts a dual-threshold endpoint detection method.

3. The method for predicting mechanical failure of parking equipment based on noise characteristics according to claim 1, characterized in that: In step S3, feature extraction of the resonance peak frequency characteristic parameter includes the following steps: A1. Frame and window the real-time mechanical noise signal or the fault-free mechanical noise signal, and then perform fast Fourier transform to obtain the short-time spectrum of the noise signal; A2. Take the logarithm of the short-time spectrum of the noise signal and perform an inverse discrete Fourier transform to obtain the cepstrum of the noise signal; A3. Obtain the cepstrum of the low-frequency part and perform fast Fourier transform to obtain a logarithmic modulus function. The peak value of the logarithmic spectrum of the logarithmic modulus function is the resonance peak frequency characteristic parameter.

4. The method for predicting mechanical failure of parking equipment based on noise characteristics according to claim 1, characterized in that: In step S3, the feature extraction of Mel-frequency cepstral coefficients includes the following steps: B1. Sampling, quantizing, and endpoint detecting the real-time mechanical noise signal or the fault-free mechanical noise signal to obtain a time domain signal; B2. Convert the time domain signal into a frequency domain signal through discrete Fourier transform or fast Fourier transform; B3, converting the frequency domain signal into a Mel spectrum through a Mel frequency filter bank; B4. Perform logarithmic energy processing on the Mel spectrum to obtain a logarithmic spectrum; B5. Perform discrete cosine transform on the logarithmic spectrum to obtain the Mel-frequency cepstral coefficients.

5. The method for predicting mechanical failure of parking equipment based on noise characteristics according to claim 1, characterized in that: The real-time mechanical noise signal is collected by a high-sensitivity pickup.

6. A parking equipment mechanical failure prediction system based on noise characteristics, characterized in that: The method comprises a processor and a memory, wherein the processor calls a program stored in the memory to implement the method for predicting mechanical failures of parking equipment based on noise characteristics as claimed in any one of claims 1 to 5.

Images