Soft measurement of ball mill load based on multi-classifier ensemble modelling and multi-sensor fusion with improved evidence combination

Self Cite

In actual industrial production, the labeled sample data of the ball mill is difficult to be obtained under variable working condition. Aiming at realizing the soft measurement of ball mill load under variable working conditions, a joint discriminative high-order moment alignment network (JDMAN) is proposed based on the deep transfer learning in this paper. In this method, the discriminative features were learned through jointly training the labeled samples belonging to source domain and unlabeled samples belonging to target domain. Simultaneously, the features learned by deep convolution network were aligned and clustered through central moment discrepancy (CMD) and center loss to accomplish transfer. The comparison with other transfer methods indicate that the proposed JDMAN can efficaciously promote the accuracy of soft measurement under variable working conditions.

Section: Experimental Data Acquisition From the Ball Millmentioning

confidence: 99%

Soft measurement of ball mill load under variable working conditions based on deep transfer learning

Huang

Guo

Sang

et al. 2022

Self Cite

“…Under high-speed and heavy-load conditions, the load acting on the rolling bearing causes flexible deformation of the raceway and rollers. It will lead to scratches, wear, fatigue spalling and other faults in the rolling bearings [1], affecting the accuracy, reliability, and service life of the aero-engine transmission 3 These authors contributed equally. * Author to whom any correspondence should be addressed.…”

Section: Introductionmentioning

confidence: 99%

Measurement of bearing reaction force by distributed strain variation

Wang

Kong

Zhang³

et al. 2023

Bearing is an important transmission component of aero-engine under high-speed and heavy-load conditions. To design a roller bearing with high loading capacity and reliability, it is essential to pay attention to the relationship between external load state (reaction force) and internal load distribution (load distribution). Therefore, a measurement method for bearing reaction force is proposed in this study via load distribution and radial basis function neural network. Different from conventional static reaction force measurement methods, both of direction and magnitude of the reaction force were considered in the proposed method without modifications to bearing. Firstly, an experimental system was designed to investigate the load distribution in a roller bearing under different reaction forces using strain variation measurement. Then, finite element analysis was conducted and simulation results of the strain variations at three interested points match well with the experimental measurements. Finally, radial basis function neural network with strong nonlinear fitting ability is applied to construct the mapping relationship between strain variation and reaction force. The results show that the proposed method can predict the reaction force with high accuracy based on the strain variation at three measuring points.

“…Due to the increasingly stringent requirements of environmental protection, energy saving and the economy, the monitoring and control of industrial processes is becoming more and more important [1,2]. Real-time and accurate measurement of key quality variables is an important way to achieve effective monitoring of industrial processes [3,4].…”

Section: Introductionmentioning

confidence: 99%

Dynamic-static collaborative strategy for industrial data modeling based on hierarchical deep networks

Peng¹,

Wang²,

Liu³

et al. 2022

With the rapid development of industrial processes, the complex nonlinear dynamic features of process data have created great challenges for deep learning models. However, existing deep learning models such as stacked autoencoder (SAE) mainly focus on capturing static data feature information while ignoring the extraction of dynamic data evolution pattern. To combat this issue, this paper proposes a novel deep learning model based on dynamic stacked autoencoder (DSAE) to collaboratively learn deep static and dynamic features of process data. Firstly, the sliding window technology is utilized to obtain temporal nearest neighbor samples within each time window. Then, multiple stacked autoencoder modules are utilized hierarchically to extract static features in each window. After that, the proposed dynamics feature extraction module is exploited to extract the local dynamics information of the data within each window. Finally, the learned dynamic and static features are collaboratively fused to build a soft sensor model for quality prediction tasks. To validate the superiority of the proposed model, it is applied to the simulation experiments of tobacco drying process and hydrocracking process. The experimental results show that the performance of the proposed model performs better than other methods.