基于故障特征稀疏表示的齿轮箱多故障模式智能分类识别方法-齿轮多故障模式实验数据集

针对风电机组工作过程中转速变化频繁造成诊断精度差的难题，开展时变工况下的齿轮箱多故障模式识别方法研究。我们开展基于稀疏字典学习的多故障模式识别方法研究。对于多故障分类问题，常规的字典学习算法是建立的单一浅层字典结构，这样的结构学习能力不强，当训练集信号不足时，学习到的字典模型准确性不高。我们提出了深度共享字典学习算法(DSDL)，它是基于多层字典层和稀疏编码层的结构来学习每一类信号中最具代表性的结构，其中每一层字典是一类共享字典，可以同时学习到多类信号中的特有成分和共享成分，有效地提高了字典的学习能力，可以解决训练样本匮乏导致学习的模型不准确的问题。本数据集通过实验台架模拟齿轮实际工作中经常可能出现的故障类型，我们对高速齿轮一共加工了五种齿轮故障类型：点蚀、断齿、裂纹、磨损、缺齿。试验中，用两个加速度传感器采集高速齿轮和低速齿轮的振动信号。数据包含6个mat文件，每一个mat文件包含两个变量，（1）fs：信号的采样率；（2）DATA：一共有4行，第一行对应高速齿轮位置的通道1，第二行为空信号，第三行对应低速齿轮位置的通道2，第四行对应速度信号。

Aiming at the challenge of poor diagnostic accuracy caused by frequent rotational speed fluctuations during wind turbine operation, we conducted research on multi-fault pattern recognition for gearboxes under time-varying operating conditions, with a focus on multi-fault pattern recognition methods based on sparse dictionary learning. For multi-fault classification tasks, conventional dictionary learning algorithms employ a single shallow dictionary structure, which exhibits limited learning capability. When training signals are insufficient, the learned dictionary model often has low accuracy. We propose a Deep Shared Dictionary Learning (DSDL) algorithm, which adopts a structure consisting of multi-layer dictionary layers and sparse coding layers to learn the most representative structures for each type of signal. Each dictionary layer functions as a shared dictionary for a specific class, enabling simultaneous extraction of both unique and shared components across multiple signal categories, effectively boosting the dictionary's learning capability and resolving the issue of inaccurate models resulting from insufficient training samples. This dataset simulates common fault scenarios in actual gear operation using an experimental test rig. We manufactured five distinct gear fault types for the high-speed gear: pitting, tooth breakage, crack, wear, and missing tooth. During the experiments, two acceleration sensors were deployed to collect vibration signals from the high-speed and low-speed gears. The dataset contains 6 MAT files, each containing two variables: (1) fs: the sampling rate of the signals; (2) DATA: with a total of 4 rows. The first row corresponds to Channel 1 at the high-speed gear position, the second row is an empty signal, the third row corresponds to Channel 2 at the low-speed gear position, and the fourth row corresponds to the rotational speed signal.

搜集汇总

背景与挑战

背景概述

该数据集包含五种齿轮故障类型（点蚀、断齿、裂纹、磨损、缺齿）的振动信号数据，用于研究基于稀疏字典学习的齿轮箱多故障模式智能分类识别方法。数据以mat文件形式存储，包含采样率和振动信号，适用于训练和验证故障诊断模型。