A feature extraction method based on composite multi-scale permutation entropy and Laplacian score for shearer cutting state recognition

Highlights

• A novel state identification method based on CMPE, LS and FOA-SVM is proposed.

• CMPE is proposed to avert the drawbacks existing in PE and MPE.

• LS is employed to achieve the selection of important and sensitive features.

• Simulation and experimental analysis proved the effectivity and superiority.

Abstract

In the field of coal mining, accurate cutting state recognition of shearer is the basis and premise for achieving automatic coal cutting. In this paper, a novel shearer cutting state recognition method is presented based on composite multi-scale permutation (CMPE), Laplacian score (LS) and fly optimization algorithm-based support vector machine classification (FOA-SVM). CMPE is proposed to overcome the shortcomings of MPE and can extract the hidden state characteristics from the vibration signals of shearer rocker arm. Some simulations are provided to select the appropriate parameter settings and prove the superiority of CMPE to MPE. In addition, LS algorithm is employed to sort the extracted features over different scales according to their importance and the sensitive feature combinations can be generated scientifically. The FOA-SVM classifier is constructed to achieve intelligent recognition of shearer cutting state. Finally, some experiments are presented and the comparison results indicated that the proposed method can realize the recognition of shearer cutting state with higher accuracy than the existing methods.

Introduction

Safe and efficient mining of coal resources is the basic guarantee to realize the unmanned fully mechanized mining face. As one piece of the core equipment of coal mining equipment, the self-adaptive cutting level of shearer is directly related to the safety and efficiency of coal mining face [2], then affects the construction of intelligent coal mine [1]. Accurate recognition of shearer cutting state is the theoretical basis for realizing adaptive cutting, and has attracted more and more attentions.

In order to improve the automation level of shearer, scholars at home and abroad mainly focus on two aspects: one is coal-rock interface recognition technology and another is memory cutting technology. In view of the first research direction, scholars have carried out a large number of theoretical and experimental studies, and put forward more than 20 methods, among which the representative methods include gamma ray detection method [7], infrared detection method [6], vibration detection method [4], image processing method [3], vibration detection method [5] and so on. However, the above methods can only identify the interface between coal seam and rock, which cannot meet the actual demand, and the application effect is not ideal for the complex geological conditions of coal seam. Memory cutting method has been applied in some mines to achieve height detection of shearer cutting drum [8] and some methods have been presented to improve the effectiveness of memory cutting, such as hidden Markov model [9] and fuzzy control theory [10]. However, when the distribution of coal seam changes suddenly, the current cutting state of shearer cannot be recognized effectively. And the position of shearer must be adjusted and controlled by manual means, which has poor adaptability and cannot truly realize the automatic cutting of shearer. In this context, the vibration signals are introduced for further analysis, which has been widely used in the field of engineering. In the process of coal cutting, the roller on the rocker arm is in direct contact with the coal wall. When the roller cuts the coal seam with different properties, the rocker arm will inevitably represent different vibration characteristics. Therefore, this paper analyzes the vibration signal of the rocker arm, and then realizes the accurate recognition of shearer cutting state.

In fact, the kernel of using vibration signals to identify the cutting state of shearer is to extract state feature information from vibration signals. As the poor working condition of shearer, the vibration signals of rocker arm are very complex and will show a certain degree of nonlinear and non-stationary characteristics. This signal peculiarity will increase the difficulty of feature extraction from vibration signals of rocker arm. Therefore, traditional linear and stationary analysis methods is no longer suitable for shearer vibration signals. In recent years, many nonlinear dynamic methods have been widely used in the field of mechanical fault diagnosis, such as approximate entropy (AE) [11], sample entropy (SE) [12], fuzzy entropy (FE) [13], permutation entropy (PE) [14], multi-scale entropy (ME) [15], and multi-scale permutation entropy (MPE) [16]. These methods can extract the nonlinear state feature hidden in the vibration signals, which cannot be extracted by the linear analysis methods. For example, in [17], Dou et al. proposed a new method based on MSE and SVM to diagnose the fault in high voltage circuit breakers. In Ref. [18], a fault diagnosis method based on fast-variational mode decomposition, MPE and Gustafson-Kessel fuzzy clustering was proposed for rolling bearing by Chen et al. In Ref. [19], a new rolling bearing fault diagnosis method based on MPE, principal component analysis and SVM was presented by Zheng et al. and the experiment results indicated the feasibility and effectiveness.

Among these nonlinear analysis methods, PE is a new measurement method for randomness and dynamic mutation of time series. It possesses the advantages of simple calculation and strong anti-noise ability, and has achieved good results in electromyographic signal processing [20], heart rate signal processing [21] and temperature complexity [22]. In addition, some scholars have applied PE to feature extraction of vibration signals of rotating machinery. For example, Zhang et al. [23] provided a feature extraction method by using PE and ensemble empirical mode decomposition, and proposed a novel hybrid model for fault detection and classification of motor bearing. Yan et al. [24] developed a new scheme based on improved variational mode decomposition and instantaneous energy distribution-permutation entropy to recognize fault category of the rolling bearing. The results show that permutation entropy can effectively detect and amplify the dynamic changes of vibration signals, and also can characterize the working conditions under different states. However, PE only detects the randomness and dynamic mutation of time series on a single scale, and cannot fully extract the hidden state characteristics. To overcome this defect, Aziz et al. [16] proposed the concept of multi-scale permutation entropy to measure the complexity and randomness of time series at different scales. On this basis, many literatures have reported the use of MPE to extract fault features from vibration signals of rotating machinery, and then to achieve fault diagnosis, such as the combination of MPE and least squares SVM [25], and the combination of local mean decomposition, MPE and SVM based binary tree [26]. Due to the complexity of shearer cutting system, the vibration signal of rocker arm not only contains important information on a single scale, but also implicate many state information on other scales. Therefore, multi-scale analysis of vibration signal is an effective method for identifying shearer cutting state.

Although many approaches for improving the automation level of shearer and several nonlinear analysis methods for signals have been proposed in above literatures, they have some common disadvantages summarized as follows. Firstly, the serious problem of coal-rock interface recognition technology and memory cutting technology is not being able to adapt to complex geological conditions of coal seam and the actual application affect is not ideal. Secondly, for most entropy-based nonlinear analysis methods, the calculation of entropy is overly dependent on the length of time series and do not comprehensively consider all the data information in the sequence. In this paper, the vibration signals of shearer rocker arm are used for further analysis. Furthermore, a new nonlinear method called composite multi-scale permutation entropy (CMPE) is put forward for extracting features from vibration signals over different scales.

After feature extraction, a higher-dimensional set of feature samples can be obtained. However, not all features are conducive to the identification of shearer cutting state, and high-dimensional samples are prone to dimensional disaster and reduce the performance of state recognition. In this paper, the Laplacian score (LS) algorithm [27] is employed to sort all features according to their importance and then a new sensitive feature set which contains the main state information can be generated. In order to achieve intelligent recognition of cutting state, fly optimization algorithm-based support vector machine classification (FOA-SVM), as an intelligent classifier, is used in this paper, which possesses better performance in the application of state identification [28] and fault diagnosis [29]. Finally, some experiments are presented and the comparison results verify the feasibility and superiority of proposed method.

The rest of this paper is organized as follows. Section 2 describes the basic theory of PE and MPE algorithms. The CMPE algorithm is presented in Section 3. The main recognition process of proposed method is given in Section 4. The experimental verification is described in Section 5 and the conclusions are drawn in Section 6.