采煤机作为煤矿开采的主要设备，一旦发生故障，不仅严重影响开采效率，还可能导致安全隐患。因此，对采煤机进行高效的故障诊断和及时维修显得至关重要。在智能故障诊断领域，由于采煤机常处于恶劣的工作环境中，容易导致故障数据的丢失或误差，从而加剧了故障数据的不平衡性，进而影响诊断准确性。此外，设计轻量化模型有助于提高故障诊断的实时性和适应性。因此，解决数据不平衡问题并发展轻量化模型，对提高故障诊断的性能具有重要意义。针对以上问题，首先探讨了采煤机滚动轴承的结构和故障模式，并对其故障机理进行了分析。然后，对轴承的一维振动信号进行了时频分析，并通过基于二维时频图像的数据增强方法解决了数据不平衡问题，并设计了一个轻量化的卷积神经网络模型进行故障诊断。最终开发出一套采煤机自适应智能诊断监测系统，具体研究如下：

（1）针对采煤机滚动轴承振动数据，对滚动轴承的振动信号分别应用了短时傅里叶变换（Short-Time Fourier Transform, STFT）和连续小波变换（Continuous Wavelet Transform, CWT），生成了两种时频数据集，比较了在不同噪声水平和工况变化下，两种数据集的故障诊断准确率。

（2）为解决采煤机滚动轴承数据不平衡的问题，在带梯度惩罚的Wasserstein生成对抗网络（Wasserstein Generative Adversarial Network with Gradient Penalty, WGAN-GP）的基础上引入了自注意力机制（Self-Attention, SA），谱归一化（Spectral Normalization, SN）和双时间尺度更新规则（Two Time-scale Update Rule, TTRU），构建了基于IWGAN-GP的采煤机滚动轴承故障时频数据增强方法。最后，通过IS（Inception Score）、FID（Frechet Inception Distance）和sFID（Spatial Frechet Inception Distance）指标以及生成样本分布的可视化来评估生成的时频图质量。

（3）鉴于采煤机工作环境的特殊性，需要及时更新故障诊断模型以确保准确率和时效性。在视觉几何群网络（Visual Geometry Group Network-16，VGG16）的基础上，引入Ghost模块和全局平均池化（Global Average Pooling，GAP）进行轻量化改进，此外，采用CBAM注意力机制和带重启的余弦退火衰减法以提升VGG16模型性能，构建了CBAM-VGG16轻量化网络。结合之前提出的数据增强方法，提出了一种基于轻量化模型和数据增强的滚动轴承故障诊断方法。

（4）考虑到采煤机的工作环境，本文使用PyQt5框架和MySQL数据库，利用Python语言开发了采煤机自适应智能诊断监测系统。该系统集成了本文提出的IWGAN-GP模型和CBAM-VGG16模型，旨在提升采煤机效率与安全，实现精确的故障诊断和预测，减少停机事故。此外，本系统的自适应更新功能和辅助功能界面，进一步满足用户需求，确保了使用的高效性与便捷性。

Shearer as the main equipment of coal mining, once the failure, not only seriously affect the mining efficiency, but also may lead to safety risks. Therefore, it is very important to carry out efficient fault diagnosis and timely maintenance of shearer. In the field of intelligent fault diagnosis, because the shearer is often in a bad working environment, it is easy to cause the loss or error of fault data, which aggravates the imbalance of fault data, and further affects the diagnostic accuracy. In addition, the design of lightweight model is helpful to improve the real-time and adaptability of fault diagnosis. Therefore, solving the problem of data imbalance and developing lightweight models are of great significance to improve the performance of fault diagnosis. Aiming at the above problems, this paper first discusses the structure and failure mode of shearer rolling bearing, and analyzes its failure mechanism. Then, the one-dimensional vibration signal is analyzed, and the data imbalance problem is solved by data enhancement method based on two-dimensional time-frequency image, and a lightweight convolutional neural network model is designed for fault diagnosis. Finally developed a shearer adaptive intelligent diagnosis and monitoring system, the specific research is as follows:

(1) The vibration data of the shearer's rolling bearing is analyzed using the Short-Time Fourier Transform (STFT) and Continuous Wavelet Transform (CWT). Two sets of time-frequency data are generated, and the fault diagnosis accuracy of these two data sets is compared under different noise levels and working conditions.

(2) To address the issue of unbalanced data in shearer rolling bearing, we propose the use of Wasserstein Generative Adversarial Network with Gradient Penalty (WGAN-GP), Self-Attention mechanism (SA), Spectral Normalization (SN), and Two Time-scale Update Rule (TTRU). An enhancement method for fault frequency data of shearer rolling bearing is constructed based on IWGAN-GP. The quality of the generated time-frequency plots is evaluated using Inception Score (IS), Frechet Inception Distance (FID), and Spatial Frechet Inception Distance (sFID) metrics, as well as visualization of the generated sample distribution.

(3) In view of the particularity of the shearer's working environment, it is necessary to update the fault diagnosis model in time to ensure accuracy and timeliness. On the basis of Visual Geometry Group Network (VGG16), Ghost module and Global Average Pooling (GAP) are introduced for lightweight improvement. In addition, The CBAM-VGG16 lightweight network is constructed by using CBAM attention mechanism and cosine annealing attenuation method with restart to improve the performance of VGG16 model. A new fault diagnosis method of rolling bearing based on lightweight model and data enhancement is proposed in this paper.

(4) Considering the working environment of shearer, this paper uses PyQt5 framework and MySQL database, and uses Python language to develop shearer adaptive intelligent diagnosis and monitoring system. This system integrates the IWGAN-GP model and CBAM-VGG16 model proposed in this paper, aiming to improve the efficiency and safety of shearer, achieve accurate fault diagnosis and prediction, and reduce downtime accidents. In addition, the adaptive updating function and auxiliary function interface of the system further meet the needs of users and ensure the efficiency and convenience of use.