Search

Tool wear prediction can ensure product quality and production efficiency during manufacturing. Although traditional methods have achieved some success, they often face accuracy and real-time performance limitations. The current study combines multi-channel 1D convolutional neural networks (1D-CNNs) with temporal convolutional networks (TCNs) to enhance the precision and efficiency of tool wear prediction. A multi-channel 1D-CNN architecture is constructed to extract features from multi-source data. Additionally, a TCN is utilized for time series analysis to establish long-term dependencies and achieve more accurate predictions. Moreover, considering the parallel computation of the designed architecture, the computational efficiency is significantly improved. The experimental results reveal the performance of the established model in forecasting tool wear and its superiority to the existing studies in all relevant evaluation indices.

Tool wear (TW) is the gradual deterioration and loss of cutting edges due to continuous cutting operations in real production scenarios. This wear can affect the quality of the cut, increase production costs, reduce workpiece accuracy, and lead to sudden tool breakage, affecting productivity and safety. Nevertheless, since conventional tool wear monitoring (TWM) approaches often employ complex physical models and empirical rules, their application to complex and non-linear manufacturing processes is challenging. As a result, this study presents a TWM model using a convolutional neural network (CNN), an Informer encoder, and bidirectional long short-term memory (BiLSTM). First, local feature extraction is performed on the input multi-sensor signals using CNN. Then, the Informer encoder deals with long-term time dependencies and captures global time features. Finally, BiLSTM captures the time dependency in the data and outputs the predicted tool wear state through the fully connected layer. The experimental results show that the proposed TWM model achieves a prediction accuracy of 99%. It is able to meet the TWM accuracy requirements of real production needs. Moreover, this method also has good interpretability, which can help to understand the critical tool wear factors.

Flame recognition is an important technique in firefighting, but existing image flame-detection methods are slow, low in accuracy, and cannot accurately identify small flame areas. Current detection technology struggles to satisfy the real-time detection requirements of firefighting drones at fire scenes. To improve this situation, we developed a YOLOv5-based real-time flame-detection algorithm. This algorithm can detect flames quickly and accurately. The main improvements are: (1) The embedded coordinate attention mechanism helps the model more precisely find and detect the target of interest. (2) We advanced the detection layer for small targets to enhance the model’s associated identification ability. (3) We introduced a novel loss function, α-IoU, and improved the accuracy of the regression results. (4) We combined the model with transfer learning to improve its accuracy. The experimental results indicate that the enhanced YOLOv5′s mAP can reach 96.6%, 5.4% higher than the original. The model needed 0.0177 s to identify a single image, demonstrating its efficiency. In summary, the enhanced YOLOv5 network model’s overall efficiency is superior to that of the original algorithm and existing mainstream identification approaches.

Herein, to accurately predict tool wear, we proposed a new deep learning network—that is, the IE-Bi-LSTM—based on an informer encoder and bi-directional long short-term memory. The IE-Bi-LSTM uses the encoder part of the informer model to capture connections globally and to extract long feature sequences with rich information from multichannel sensors. In contrast to methods using CNN and RNN, this model could achieve remote feature extraction and the parallel computation of long-sequence-dependent features. The informer encoder adopts the attention distillation layer to increase computational efficiency, thereby lowering the attention computational overhead in comparison to that of a transformer encoder. To better collect location information while maintaining serialization properties, a bi-directional long short-term memory (Bi-LSTM) network was employed. After the fully connected layer, the tool-wear prediction value was generated. After data augmentation, the PHM2010 basic dataset was used to check the effectiveness of the model. A comparison test revealed that the model could learn more full features and had a strong prediction accuracy after hyperparameter tweaking. An ablation experiment was also carried out to demonstrate the efficacy of the improved model module.

SCoralNet (based on Faster R-CNN) is a new underwater coral detection framework that has been proposed to automatically localize and identify distinct coral species in images, allowing for fast and detailed annotation. Monitoring the coverage and abundance of underwater corals typically involves the annotation and processing of large amounts of underwater coral images. However, manually annotating a large number of images is time-consuming and labor-intensive, and CNN classifiers only provide simple classification annotations without capturing the images' finer details. SCoralNet's detection performance is improved by incorporating dilated convolutions into the backbone network. To successfully capture multi-scale and multi-level information from coral targets, a neck network called NASFPN is placed between the backbone and the detecting head. Seesaw Loss is used to reduce the impact of the dataset's long-tailed distribution on SCoralNet's classifier accuracy. CIoU loss is used to optimize the bounding box regression method. During inference, Soft-NMS is applied to suppress redundant coral detection boxes. To assess SCoralNet's effectiveness, a dataset called Coral-soft was developed using real-world photos of common soft coral species from the Sanya region of China. SCoralNet outperformed the original Faster R-CNN model on the Coral-soft dataset, with a 45.68% gain in mean average precision (mAP) and a 59.2% increase in mAP75. Furthermore, SCoralNet outperformed some advanced models in terms of overall performance. [ABSTRACT FROM AUTHOR]