1. Introduction
The shift from traditional industry to smart manufacturing envisioned by the Industry 4.0 paradigm creates opportunities to leverage recent advances in artificial intelligence (AI) to improve productivity and safety [1,2,3]. Despite these advances, handling heavy loads remains a major safety issue in various manufacturing environments. Cranes are crucial in heavy-duty handling as they can move objects weighing several tons. Cranes are designed to lift, lower and transport heavy objects with precision and efficiency, facilitating the smooth operation of various industrial processes.Handling heavy objects, on the other hand, involves inherent risks and challenges [4]. Over time, crane hoisting ropes undergo significant stress and wear.Load weight, frequency of use, environmental conditions and operating requirements can all cause hoisting cables to deteriorate [5]. It is critical to regularly monitor and inspect these ropes to ensure structural integrity and avoid potential accidents or equipment failure. Lifting cables are particularly prone to failure and must be inspected and replaced regularly.Manual inspection procedures, on the other hand, are labor-intensive, time-consuming, subjective, and often require temporary shutdowns of the production process [6].
Lifting ropes are usually made of steel wire, but sometimes synthetic fibers may be used [7]. Steel cables are susceptible to damage during use, reducing their strength and posing a safety risk. A thorough understanding of the various types of damage to wire ropes is critical for accurate assessment. These types of damage include broken wires, wear, deformation, rust, and fatigue. Fatigue, in particular, can take many forms, such as cracks, broken wires, and sagging.Proper assessment and identification of these damage types is critical to the integrity and safety of steel wire cables [8].Cable inspection and testing are critical safety measures performed at the final stages of production and pre-installation operations [9]. Detecting and repairing defects in cables and wires is critical to avoiding financial losses and ensuring user safety and well-being. Therefore, cable manufacturers and end users recognize that it is critical to perform multiple tests and inspections to reduce potential risks and ensure reliable cable performance.
Steel cable fault detection methods are generally classified as non-destructive or destructive.Non-destructive testing methods use specialized testing instruments or rely on observation of anomalies to visualize defects or anomalies within cable materials [10]. The captured signals, images or parameters are then subjected to extensive analysis, evaluation and judgment. Non-destructive testing is characterized by the ability to evaluate the internal condition of structural materials without causing any damage to their structure, performance or dimensional integrity. This ensures that the original shape and functionality of the cable is retained throughout the testing process.Metal cable inspection methods include manual inspection, automatic optical inspection (AOI) and magnetic flux leakage inspection [11]. Flux leakage testing evaluates the condition of metal cables by measuring loss of metal area (LMA). Relevant guidance such as ISO 4309 [12] Follow to ensure compliance with industry standards. This standard covers the basic principles of testing, procedures, appropriate instrumentation and calibration methods required for accurate and reliable cable testing. Following these established standards is critical to maintaining consistency and quality in the testing process.The material’s magnetic properties, magnetization level, and defect characteristics (such as depth, width, length, tilt angle, etc.) all affect the effectiveness of magnetic flux leakage inspection equipment [13].
Detecting faults in metal cable systems is critical to the safety and reliability of applications such as bridges and elevators, which face complex test environments with signal interference and noise. After signal acquisition and processing, signal analysis becomes very complicated due to the reduction in signal-to-noise ratio and detection sensitivity.Various signal processing techniques, including impulse filters, data fusion, and feature extraction, have been explored to address these challenges [14,15].Incorporating advanced mathematical techniques into physical testing of various detection methods is a catalyst for accuracy and efficiency [16].This involves employing Fourier analysis, machine learning algorithms, image reconstruction and simulation to enhance the inspection process [17,18]. These technologies provide solutions to complex challenges and advance the field of detection technology.Innovative algorithms driven by supervised and unsupervised learning methods and neural network technology [19,20,21,22,23], emerged to meet changing engineering requirements. For example, Katir et al. A new PSO-YUKI algorithm with RBF is introduced for rapid damage identification of CFRP laminates, which is superior to traditional PSO in double crack depth assessment. Mao Qinghua et al. [24] An improved decision tree support vector machine (SVM) algorithm was introduced and verified experimentally to improve the classification accuracy of metal cord conveyor belt defects.Implementation of Artificial Neural Networks (ANN) and advanced algorithms for structural health monitoring of laminated composite panels, focusing on damage localization and quantification [25]. In addition, an artificial neural network is used to predict the displacement of the composite tube at different speeds.This reflects continued advances in the field of ANN technology for structural health monitoring [26].Beyond traditional machine learning, deep learning technology [27]especially convolutional neural networks (CNN) [28], has achieved outstanding achievements in various fields including signal recognition. Piaquiglia et al. [29] Fully connected neural networks (fully convolutional networks, FCN) and one-dimensional convolutional neural networks (1D CNN) are used to analyze and classify ECG signals, while Zhang and Wu [12] Apply Deep Belief Network (DBN) to analyze sound signals and speech activity detection. Zhou Gongbo et al. [30] Demonstrates the usefulness of CNN for real-time fault detection in metal cable fault detection. At the same time, Liu Zhiliang et al. [31] A CNN-based surface defect detection method was developed, demonstrating its strong learning ability and high diagnostic accuracy in metal cable systems. Liu et al. [32] Examining wire rope defect identification using MFL signal analysis and 1D CNN, emphasizing the role of signal processing and machine learning in improving accuracy, reflecting current trends in the field. These examples highlight the critical role of data processing in enabling effective signal identification and defect detection in metallic cable systems.
K-mers play a variety of roles in bioinformatics, including quality control of sequence generation, metagenomic classification, and genome size estimation [33,34]. Meplassen et al.Describe KAT as a tool for NGS data quality control K-mer analysis, providing k value guidance and tool comparison [35]. Breitweiser et al.Introducing KrakenUniq, which combines the speed of Kraken with the uniqueness of efficiency KMer coverage assessment in metagenomics [33]. Taha et al.Extract amino acids K-mer Functions for machine learning-based quantitative antimicrobial resistance (AMR) prediction and provide model interpretation for biological insights [36].Machine learning methods based on K-mers has also shown promise in pattern recognition problems, by quantizing to a fixed length KPolymer frequency in DNA sequences [37,38]. Akaya et al. Use deep learning to evaluate sequence representations and highlight their role in model performance.They show that they retain K– Relationships are critical to achieving better results [39].Integrate KMer frequencies of deep learning methods exploit their complementary strengths to improve the accuracy of data classification.
In this study, manual techniques are employed to train and identify signal data. The signals obtained by the magnetic flux leakage detection equipment are normalized.Later, it was proposed KApply -mer frequency encoding method. Finally, Autoencoder (AE) training recognition is the reason why Autoencoder is chosen because it can identify both defect-free and defective data by simply inputting defect-free data.
4. Discussion and limitations
This study yields important insights into the effectiveness of artificial intelligence strategies in cable defect identification. We began our study by training the autoencoder using defect-free cable signals, a key choice that allowed the model to perform well at reconstructing defect-free signals, thus establishing a method for identifying defective signals based on reconstruction error values. A practical approach to defective cable signaling. This observation has great practical significance and shows that the reconstruction error value can be used as a reliable indicator for detecting cable defects.
A significant contribution of this study is the introduction of K-mer Frequency encoding method for cable data processing. Our subsequent experiments aim to evaluate its impact on the accuracy of the autoencoder model for cable defect recognition.Our comparative analysis reveals compelling results for a model using KThe -mer frequency encoding method achieved an astonishing maximum accuracy of 91%. This result is significantly better than the model trained using the original data, with a maximum accuracy of 81%.These findings highlight the significant improvement in accuracy K-mer frequency encoding enables cable signal analysis.Furthermore, our study explores the impact of repeatedly sampled data on cable defect identification models, showing smaller K value and improve model accuracy.This indicates the choice K values play a key role in enhancing the model’s ability to accurately identify cable defects.Furthermore, our comparative analysis K-mer frequency sampling indicates K The accuracy of value 4 is up to 91%, and the accuracy gradually decreases as the accuracy decreases. K value increases.This emphasizes choosing the appropriate K value in use K-mer frequency encoding method in cable signal analysis provides a promising approach to optimize cable defect identification in industrial applications.
One of the limitations of this study is that the method mainly applies to smaller datasets, and using it on larger datasets may result in the generation of larger amounts of data, which may significantly extend the training time of the model. This can pose challenges for real-world applications, where efficient training and inference times are critical. Additionally, segmenting the data too finely to accommodate a larger dataset may result in the loss of important data features, which may impact model performance. When data is scarce due to excessive data segmentation, models may suffer from reduced accuracy and generalization capabilities. Furthermore, the computational resources required to train and process large datasets should be carefully considered to ensure effective implementation in real-world applications. Despite these limitations, the proposed method shows promising results on smaller datasets and lays the foundation for further exploration and optimization in larger datasets and real-life scenarios.