Fault diagnosis method for wheeled traveling mechanism of gantry crane

Abstract

The invention relates to the technical field of gantry cranes, in particular to a gantry crane wheel type walking mechanism fault diagnosis method. The method comprises the following steps: acquiring vibration data, sound data and image data of a to-be-diagnosed walking mechanism, and performing multi-source data alignment by adopting timestamps; preprocessing the acquired data, and separating environmental noise from the sound data; comparing the environment noise data with a dynamic threshold value, and dividing the environment noise data into low noise, medium noise and high noise; for different noise levels, different models are adopted to complete fault diagnosis.

Description

Technical Field

[0001] The present invention relates to the technical field of gantry cranes, and in particular to a method for diagnosing faults of a wheeled traveling mechanism of a gantry crane. Background Art

[0002] Gantry cranes are key transport equipment in large-scale logistics, construction, and industrial fields. Their wheeled traveling mechanisms are key components that support the movement of the entire machine. Steel wires, as important elements connecting the movable pulley, fixed pulley, and drive system, not only bear enormous tension but also frequently bend and stretch, making them prone to wear, breakage, and other faults. This seriously affects the safety and stability of the crane. Gantry cranes often operate in complex environments, such as ports and construction sites. The slenderness and flexibility of the steel wires make it extremely difficult to directly install sensors for monitoring, and they are easily disturbed by environmental noise, affecting monitoring accuracy. Therefore, the present invention proposes a method of installing sensors on the body of the wheeled traveling mechanism and on the movable and fixed pulleys, and indirectly monitoring the vibration, sound, and image information of the steel wires to achieve real-time monitoring of wheeled traveling mechanism faults. Furthermore, different weights are assigned to the three types of information based on different working environments to improve the accuracy of judging the gantry crane traveling mechanism.

[0003] Bridge cranes are a common and important type of large-scale tool equipment that enables mechanization and automation of production processes. As core equipment, bridge cranes are responsible for the lifting, transportation, and loading and unloading of construction materials and large equipment between construction sites. They are particularly indispensable for high-altitude operations and the installation of large components. With the rapid development of my country's manufacturing, mining, and construction industries, bridge cranes are becoming increasingly important in construction projects, providing efficient handling solutions for material handling and loading and unloading, production line collaboration, and optimized space utilization.

[0004] In construction projects, bridge cranes often undertake a large number of high-frequency and high-load operations, and often face harsh and complex construction environments such as high temperatures, low temperatures, and strong winds. Due to the importance and frequent use of bridge cranes in construction projects, the risk of equipment safety hazards and failures also increases. In particular, the reliability and safety of the hoisting mechanism are crucial to the quality of construction projects and the safety of construction workers. Therefore, ensuring the stable and efficient operation of bridge cranes at construction sites and ensuring the safety and reliability of hoisting and handling operations have become important factors in promoting production efficiency and safety performance in the construction industry. Summary of the Invention

[0005] The present invention discloses a method for diagnosing a fault of a wheeled traveling mechanism of a gantry crane, and the specific method is as follows:

[0006] Obtain vibration data, sound data, and image data of the walking mechanism to be diagnosed, and use timestamps to align multi-source data;

[0007] Preprocessing the acquired data to separate environmental noise from the sound data;

[0008] Compare the environmental noise data with the dynamic threshold and classify it into low noise, medium noise and high noise;

[0009] For low-noise classification, the unprocessed sound data, vibration data, and image data are fused and input into the low-noise diagnosis model to complete the walking mechanism fault diagnosis; for medium-noise classification, the processed sound data, vibration data, and image data are fused and input into the medium-noise diagnosis model to complete the walking mechanism fault diagnosis; for high-noise classification, the vibration data and image data are fused and input into the high-noise diagnosis model to complete the walking mechanism fault diagnosis.

[0010] Furthermore, the vibration data is obtained as follows:

[0011] The sensor is installed on the bearing seat of the brake pulley and fixed pulley of the walking mechanism to be diagnosed, and the vibration data of the wire is collected, including the amplitude mean, amplitude variance, amplitude peak, frequency mean, frequency variance and frequency peak.

[0012] Furthermore, the method for obtaining the sound data is as follows:

[0013] The sound data generated during the movement of the steel wire of the walking mechanism to be diagnosed is collected through sensors, including sound intensity and sound frequency.

[0014] Furthermore, the sound data is preprocessed as follows:

[0015] Divide the continuous time domain data into several overlapping short time frames to complete the framing operation;

[0016] Each frame of short-time frame data is multiplied by the window function to complete the windowing operation;

[0017] Each frame of data is transformed by fast Fourier transform to convert time domain data into frequency domain data;

[0018] Subtract the noise frequency from the noisy spectrum using the formula:

[0019] S(k,m)=max{∣Y(k,m)∣-α∣N(k,m)∣,β∣N(k,m)∣}

[0020] Where S(k,m) is the estimated noise spectrum, Y(k,m) is the sound of the walking mechanism, and N(km) is the ambient noise. α is the over-reduction coefficient, and β is the lower limit coefficient.

[0021] Perform inverse transform on the modified spectrum, splice the processed data frames together, complete the recovery of amplitude and concomitant, and form the sound data after removing the environmental noise;

[0022] The sound data after removing the environmental noise is filtered and enhanced. The specific formula is as follows:

[0023]

[0024] Where E(j) is the energy of the Mel filter bank.

[0025] Furthermore, the vibration data is preprocessed as follows:

[0026] Set the candidate set of smoothing coefficient α;

[0027] For each candidate value of the smoothing coefficient α, perform exponential smoothing prediction to calculate the performance index and select the optimal smoothing coefficient α;

[0028] The vibration data is smoothed using the optimal smoothing coefficient.

[0029] Furthermore, we train low, medium, and high noise diagnosis models as follows:

[0030] The information entropy of each data source is redistributed based on the entropy weight method. The specific formula is:

[0031]

[0032] Among them, p ij is the probability distribution of the i-th type of data in the j-th feature dimension;

[0033] Dynamically calculate the weights of vibration data, sound data, and image data. The specific formula is:

[0034]

[0035] Among them, m is the data source type number, H i is the information entropy of the i-th data source;

[0036] The data of different environmental noise levels are divided into training set, validation set and test set, and low, medium and high noise diagnosis models are trained respectively.

[0037] Furthermore, the dynamic threshold is determined by the particle swarm optimization algorithm, and the specific method is as follows:

[0038] The noise energy calculation formula is defined as follows:

[0039]

[0040] Where T is the time window, N(t) is the time domain sampling value of the noise signal;

[0041] Construct the fitness function of the optimization objective. The specific formula is:

[0042]

[0043] Initialize the low threshold E1 and the high threshold E2, θ = [E1, E2];

[0044] Determine the search range of the pants group optimization algorithm: E1∈[μ-3σ,μ-σ], E2∈[μ+σ,μ+3σ] Initially force E1<E2 and ensure that: E low <E mid <E high They are low noise, medium noise and high noise;

[0045] Initialize particle swarm;

[0046] Calculate the fitness of each particle;

[0047] Update particle velocity;

[0048] Iterative calculation and update to select the optimal dynamic threshold.

[0049] Furthermore, the dynamic threshold is determined by the particle swarm optimization algorithm and then dynamically updated in real time. The specific method is as follows:

[0050] Calculate the current window noise energy in real time. The specific formula is as follows:

[0051]

[0052] N(t) is the time domain sampling value of the noise signal based on the time window T;

[0053] Calculate the noise volatility. The specific formula is as follows:

[0054]

[0055] When energy exceeds the limit and frequency domain anomalies occur simultaneously, the following formula is used to correct the threshold:

[0056] E new =E2+γ·(E current -E2)

[0057] Among them, Ecurrent is the noise energy value of the current time window, and γ is the learning rate;

[0058] After dynamic real-time updating, the change in classification accuracy is calculated; if the accuracy decreases, the learning rate γ is reduced; if the accuracy increases, the current learning rate γ is maintained.

[0059] Due to the adoption of the above technical solution, the present invention has the following beneficial effects:

[0060] 1. The present invention effectively overcomes the limitations of a single data source in complex environments through multimodal fusion of vibration data, sound data, and image data, combined with a dynamic weight allocation mechanism, and significantly improves the accuracy and robustness of fault diagnosis.

[0061] 2. The present invention adopts the particle swarm optimization algorithm to adjust the noise threshold in real time, and dynamically switches the diagnostic model in combination with the ambient noise level, thus solving the problem of insufficient sensitivity of traditional methods in high-noise and high-fluctuation environments, and ensuring the reliability of diagnostic results under different working conditions.

[0062] 3. The present invention indirectly monitors the state of the steel wire through non-invasive sensors, avoiding damage to the equipment structure caused by directly installing sensors, while reducing the frequency of manual inspections and lowering equipment maintenance costs.

[0063] Other advantages, objects, and features of the present invention will be described in part in the following description and, in part, will be apparent to those skilled in the art upon examination of the following description or may be learned from practice of the present invention. The objects and other advantages of the present invention may be realized and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] The accompanying drawings of the present invention are described below.

[0065] Figure 1 It is a schematic diagram of the overall process of the present invention.

[0066] Figure 2 Schematic diagram of the process for determining dynamic thresholds. DETAILED DESCRIPTION

[0067] The present invention will be further described below with reference to the accompanying drawings and examples.

[0068] A method for diagnosing faults of wheeled traveling mechanisms of gantry cranes, such as Figure 1 As shown, the specific steps are as follows.

[0069] S1. Obtain vibration data, sound data, and image data of the walking mechanism to be diagnosed, and use timestamps to align multi-source data.

[0070] In step S1, vibration data is collected indirectly from the bearing housings attached to the movable and fixed pulleys, collecting statistical characteristics of the wire's vibration data, including amplitude, frequency, and other parameters. This vibration data can reveal faults such as wear, looseness, and imbalance in the gantry crane's wheeled traveling mechanism's wire.

[0071] Sound data: This tool collects sound data generated by the movement of the steel wire of the gantry crane's wheeled traveling mechanism, including sound intensity and frequency distribution. The collected sound data is then de-noised and separated to generate the steel wire sound data.

[0072] Image data: Image information of the steel wire of the wheeled traveling mechanism of the gantry crane and its surrounding environment.

[0073] S2. Preprocess the acquired data to separate environmental noise from the sound data.

[0074] Preprocess the sound data as follows:

[0075] Divide the continuous time domain data into several overlapping short time frames to complete the framing operation;

[0076] Each frame of short-time frame data is multiplied by the window function to complete the windowing operation;

[0077] Each frame of data is transformed by fast Fourier transform to convert time domain data into frequency domain data;

[0078] Subtract the noise frequency from the noisy spectrum using the formula:

[0079] S(k,m)=max{∣Y(k,m)∣-α∣N(k,m)∣,β∣N(k,m)∣}

[0080] Where S(k,m) is the estimated noise spectrum, Y(k,m) is the sound of the walking mechanism, and N(km) is the ambient noise. α is the over-reduction coefficient, and β is the lower limit coefficient.

[0081] Perform inverse transform on the modified spectrum, splice the processed data frames together, complete the recovery of amplitude and concomitant, and form the sound data after removing the environmental noise;

[0082] The sound data after removing the environmental noise is filtered and enhanced. The specific formula is as follows:

[0083]

[0084] Where E(j) is the energy of the Mel filter bank.

[0085] Preprocess the vibration data as follows:

[0086] Set the candidate set of smoothing coefficient α;

[0087] For each candidate value of the smoothing coefficient α, perform exponential smoothing prediction to calculate the performance index and select the optimal smoothing coefficient α;

[0088] The vibration data is smoothed using the optimal smoothing coefficient.

[0089] S3. Compare the environmental noise data with the dynamic threshold and classify it into low noise, medium noise and high noise.

[0090] The process of determining dynamic threshold is as follows Figure 2 As shown, the details are as follows:

[0091] The noise energy calculation formula is defined as follows:

[0092]

[0093] Where T is the time window, N(t) is the time domain sampling value of the noise signal;

[0094] Construct the fitness function of the optimization objective. The specific formula is:

[0095]

[0096] Initialize the low threshold E1 and the high threshold E2, θ = [E1, E2];

[0097] Determine the search range of the pants group optimization algorithm: E1∈[μ-3σ,μ-σ], E2∈[μ+σ,μ+3σ] Initially force E1<E2 and ensure that: E low <E mid <E high They are low noise, medium noise and high noise;

[0098] Initialize particle swarm;

[0099] Calculate the fitness of each particle;

[0100] Update particle velocity;

[0101] Iterative calculation and update to select the optimal dynamic threshold.

[0102] S4. For low noise classification, the unprocessed sound data, vibration data and image data are integrated and input into the low noise diagnosis model to complete the walking mechanism fault diagnosis; for medium noise classification, the processed sound data, vibration data and image data are integrated and input into the medium noise diagnosis model to complete the walking mechanism fault diagnosis; for high noise classification, the vibration data and image data are integrated and input into the high noise diagnosis model to complete the walking mechanism fault diagnosis.

[0103] In step S4, low, medium, and high noise diagnosis models are trained. The specific method is as follows:

[0104] The information entropy of each data source is redistributed based on the entropy weight method. The specific formula is:

[0105]

[0106] Among them, pij is the probability distribution of the i-th type of data in the j-th feature dimension;

[0107] Dynamically calculate the weights of vibration data, sound data, and image data. The specific formula is:

[0108]

[0109] Among them, m is the data source type number, H i is the information entropy of the i-th data source;

[0110] The data of different environmental noise levels are divided into training set, validation set and test set, and low, medium and high noise diagnosis models are trained respectively.

[0111] After the dynamic threshold is determined by the particle swarm optimization algorithm, it is dynamically updated in real time. The specific method is as follows:

[0112] Calculate the current window noise energy in real time. The specific formula is as follows:

[0113]

[0114] N(t) is the time domain sampling value of the noise signal based on the time window T;

[0115] Calculate the noise volatility. The specific formula is as follows:

[0116]

[0117] When energy exceeds the limit and frequency domain anomalies occur simultaneously, the following formula is used to correct the threshold:

[0118] E new =E2+γ·(E current -E2)

[0119] Among them, Ecurrent is the noise energy value of the current time window, and γ is the learning rate;

[0120] After dynamic real-time updating, calculate the change in classification accuracy; if the accuracy decreases, reduce the learning rate; if the accuracy increases, maintain the current learning rate.

[0121] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered by the scope of protection of the claims of the present invention.

Claims

1. A method for diagnosing a fault of a wheeled traveling mechanism of a gantry crane, characterized in that: The specific method is as follows: Obtain vibration data, sound data, and image data of the walking mechanism to be diagnosed, and use timestamps to align multi-source data; Preprocessing the acquired data to separate environmental noise from the sound data; Compare the environmental noise data with the dynamic threshold and classify it into low noise, medium noise and high noise; For low-noise classification, the unprocessed sound data, vibration data, and image data are fused and input into the low-noise diagnosis model to complete the walking mechanism fault diagnosis; for medium-noise classification, the processed sound data, vibration data, and image data are fused and input into the medium-noise diagnosis model to complete the walking mechanism fault diagnosis; for high-noise classification, the vibration data and image data are fused and input into the high-noise diagnosis model to complete the walking mechanism fault diagnosis.

2. The method for diagnosing a fault of a wheeled traveling mechanism of a gantry crane according to claim 1, wherein: The method for obtaining vibration data is as follows: The sensor is installed on the bearing seat of the brake pulley and fixed pulley of the walking mechanism to be diagnosed, and the vibration data of the wire is collected, including the amplitude mean, amplitude variance, amplitude peak, frequency mean, frequency variance and frequency peak.

3. The method for diagnosing a fault of a wheeled traveling mechanism of a gantry crane according to claim 1, wherein: The method for obtaining sound data is as follows: The sound data generated during the movement of the steel wire of the walking mechanism to be diagnosed is collected through sensors, including sound intensity and sound frequency.

4. The method for diagnosing a fault of a wheeled traveling mechanism of a gantry crane according to claim 1, wherein: Preprocess the sound data as follows: Divide the continuous time domain data into several overlapping short time frames to complete the framing operation; Each frame of short-time frame data is multiplied by the window function to complete the windowing operation; Each frame of data is transformed by fast Fourier transform to convert time domain data into frequency domain data; Subtract the noise frequency from the noisy spectrum using the formula: S(k,m)=max{∣Y(k,m)∣-α∣N(k,m)∣,β∣N(k,m)∣} Where S(k,m) is the estimated noise spectrum, Y(k,m) is the sound of the walking mechanism, and N(km) is the ambient noise. α is the over-reduction coefficient, and β is the lower limit coefficient. Perform inverse transform on the modified spectrum, splice the processed data frames together, complete the recovery of amplitude and concomitant, and form the sound data after removing the environmental noise; The sound data after removing the environmental noise is filtered and enhanced. The specific formula is as follows: Where E(j) is the energy of the Mel filter bank.

5. The method for diagnosing a fault of a wheeled traveling mechanism of a gantry crane according to claim 1, wherein: Preprocess the vibration data as follows: Set the candidate set of smoothing coefficient α; For each candidate value of the smoothing coefficient α, perform exponential smoothing prediction to calculate the performance index and select the optimal smoothing coefficient α; The vibration data is smoothed using the optimal smoothing coefficient.

6. The method for diagnosing a fault of a wheeled traveling mechanism of a gantry crane according to claim 1, wherein: Train low, medium, and high noise diagnostic models as follows: The information entropy of each data source is redistributed based on the entropy weight method. The specific formula is: Among them, p ij is the probability distribution of the i-th type of data in the j-th feature dimension; Dynamically calculate the weights of vibration data, sound data, and image data. The specific formula is: Among them, m is the data source type number, H i is the information entropy of the i-th data source; The data of different environmental noise levels are divided into training set, validation set and test set, and low, medium and high noise diagnosis models are trained respectively.

7. The method for diagnosing a fault of a wheeled traveling mechanism of a gantry crane according to claim 1, wherein: The dynamic threshold is determined by the particle swarm optimization algorithm. The specific method is as follows: The noise energy calculation formula is defined as follows: Where T is the time window, N(t) is the time domain sampling value of the noise signal; Construct the fitness function of the optimization objective. The specific formula is: Initialize the low threshold E1 and the high threshold E2, θ = [E1, E2]; Determine the search range of the pants group optimization algorithm: E1∈[μ-3σ,μ-σ], E2∈[μ+σ,μ+3σ] Initially force E1<E2 and ensure that: E low <E mid <E high They are low noise, medium noise and high noise; Initialize the particle swarm; Calculate the fitness of each particle; Update particle velocity; Iterative calculation and update to select the optimal dynamic threshold.

8. The method for diagnosing a fault of a wheeled traveling mechanism of a gantry crane according to claim 7, wherein: After the dynamic threshold is determined by the particle swarm optimization algorithm, it is dynamically updated in real time. The specific method is as follows: Calculate the current window noise energy in real time. The specific formula is as follows: N(t) is the time domain sampling value of the noise signal within the time window T; Calculate the noise volatility. The specific formula is as follows: When energy exceeds the limit and frequency domain anomalies occur simultaneously, the following formula is used to correct the threshold: E new =E2+γ·(E current -E2) Among them, Ecurrent is the noise energy value of the current time window, and γ is the learning rate; After dynamic real-time updating, the change in classification accuracy is calculated; if the accuracy decreases, the learning rate γ is reduced; if the accuracy increases, the current learning rate γ is maintained.

Images