Hydrogen is considered to be a hazardous substance. Hydrogen sensors can be used to detect the concentration of hydrogen and provide an ideal monitoring means for the safe use of hydrogen energy. Hydrogen sensors need to be highly reliable, so fault identification and diagnosis for gas sensors are of vital practical significance. However, traditional machine learning methods for fault diagnosis are based on features extracted by experts, prior knowledge requirements and the sensitivity of system changes. In this study, a new convolutional neural network (CNN) using the random forest (RF) classifier is proposed for hydrogen sensor fault diagnosis. First, the 1-D time-domain data of fault signals are converted into 2-D gray matrix images; this process does not require noise suppression and no signal information is lost. Secondly, the features of the gray matrix images are automatically extracted by using a CNN, which does not rely on expert experience. Dropout and zero-padding are used to optimize the structure of the CNN and reduce overfitting. Random forest, which is robust and has strong generalization ability, is introduced for the classification of gas sensor signal modes, in order to obtain the final diagnostic results. Finally, we design and implement a prototype hydrogen sensor array for experimental verification. The accuracy of fault diagnosis in hydrogen sensors is 100% under noisy environment with the proposed method, which is superior of CNN without RF and other methods. The results show that the proposed CNN with RF method provides a good solution for hydrogen sensor fault diagnosis. INDEX TERMS Fault diagnosis, hydrogen sensor, convolutional neural network, random forest, feature extraction.

The fault safety monitoring of hydrogen sensors is very important for their practical application. The precondition of traditional machine learning methods for sensor fault diagnosis is that enough fault data with the same distribution and feature space under the same working environment must exist. Widely used fault diagnosis methods are not suitable for real working environments because they are easily complicated by environmental conditions such as temperature, humidity, shock, and vibration. Under the influence of such complex conditions, the acquisition of sensor fault data is limited. In order to improve fault diagnosis accuracy under complex environmental conditions, a novel method of transfer learning (TL) with LeNet-5 is proposed in this paper. Firstly, LeNet-5 is applied to learn the features of the data-rich datasets of gas sensor faults in a normal environment and to adjust the parameters accordingly. The parameters of the LeNet-5 are transferred from the task in the normal environment to a task in a complex environment by using the TL method. Then, the migrated LeNet-5 is used for the fault diagnosis of gas sensors with a small amount of fault data in a complex environment. Finally, a prototype hydrogen sensor array is designed and implemented for experimental verification. The gas sensor fault diagnosis accuracy of the traditional LeNet-5 was 88.48 ± 1.04%, while the fault diagnosis accuracy of TL with LeNet-5 was 92.49 ± 1.28%. The experimental results show that the method adopted presents an excellent solution for the fault diagnosis of a hydrogen sensor using a small quantity of fault data obtained under complex environmental conditions.

In recent years, the treatment of breast cancer has advanced dramatically and neoadjuvant chemotherapy (NAC) has become a common treatment method, especially for locally advanced breast cancer. However, other than the subtype of breast cancer, no clear factor indicating sensitivity to NAC has been identified. In this study, we attempted to use artificial intelligence (AI) to predict the effect of preoperative chemotherapy from hematoxylin and eosin images of pathological tissue obtained from needle biopsies prior to chemotherapy. Application of AI to pathological images typically uses a single machine‐learning model such as support vector machines (SVMs) or deep convolutional neural networks (CNNs). However, cancer tissues are extremely diverse and learning with a realistic number of cases limits the prediction accuracy of a single model. In this study, we propose a novel pipeline system that uses three independent models each focusing on different characteristics of cancer atypia. Our system uses a CNN model to learn structural atypia from image patches and SVM and random forest models to learn nuclear atypia from fine‐grained nuclear features extracted by image analysis methods. It was able to predict the NAC response with 95.15% accuracy on a test set of 103 unseen cases. We believe that this AI pipeline system will contribute to the adoption of personalized medicine in NAC therapy for breast cancer.

The corrosion is an important problem for the service safety of oil and gas pipeline. This research focuses. This paper proposed a new prediction algorithm on corrosion prediction of gathering gas pipeline, which combined modified Support Vector Machine (SVM) with unequal interval model. Firstly, grey prediction method with unequal interval model was used to pretreatment original data because there is unequal interval problem in actual collected data of pipeline. Secondly, improved Support Vector Regression (SVR) based on Genetic Algorithm (GA) and Particle Swarm Optimization (PSO) has been proposed to resolve parameters selection problem for SVR. Finally, the corrosion prediction model of gas pipeline has been proposed which combined improved SVR and unequal interval grey prediction method. The experiment results show this algorithm could increase precision of the pipeline corrosion prediction compared with the traditional SVM. This research provides reliable basis for in-service pipeline life prediction and confirming inspecting cycle.

No abstract