Deep Learning Algorithm Generalization for Complex Industrial Systems

Dear Colleagues,

The AI implementation scenario can be compared to an iceberg. The industry is the hidden "treasure" under the horizontal surface, which is exceptionally large and and has great potential, but at the same time, it is particularly challenging to overcome. Implementing industrial intelligence is destined to be a tough and protracted battle. A good algorithm model requires massive data and arithmetic support. Still, industrial scene data are seriously lacking, and the arithmetic cost of installing GPUs is difficult for traditional enterprises to accept.   The more critical point is that the "black box" characteristic of deep learning is naturally contradictory to the pursuit of industrial manufacturing's accuracy, reliability, and interpretability. It is not easy to gain the trust of industrial enterprises. In contrast, decision trees, classification algorithms, regression analyses, and other classical machine learning algorithms are more widely used in the industrial field. In the process of industrial intelligence implementation, a headache is "one machine, one model", and industrial algorithms are difficult to generalize. Algorithm generalization is crucial and directly affects whether it can be called a product. After all, if it cannot be productized, it cannot be scaled up. However, the working conditions of industrial systems are particularly complex. For example the materials, structures, and models of tools; the performance of the processing machine; the material and structure of the workpiece; and the site environment are all different, which often leads to the creation of models only for a specific working conditions. Put into other working conditions, these models’ effects are greatly reduced. The core reason behind this is that the complexity and process threshold of the industry is extremely high. At the same time, the amount of data available for modeling is generally scarce and of low quality, and the lack of industry knowledge and mechanisms makes it difficult for data-driven models to have good generalization capabilities.

From the modeling point of view, fusing industry experts' knowledge and mechanism models into machine learning models can often reduce the required training data several times and make the models more adaptable to different environments and working conditions. From the feature perspective, extracting features with certain mechanical properties can enhance the causal properties of model judgments and significantly reduce the required computational effort. Compared with not adding mechanistic features, adding mechanistic features usually improves the accuracy of the model, only that the degree of improvement may vary from scenario to scenario. Currently, industry and scholars try to use migration learning to improve the generalization ability of models, but it is still in the exploration stage now, and it still takes time to really move toward the ground. In addition, the ability of model generalization itself is limited, and it needs to be complemented by a series of engineering means from the product dimension at this time.

This Special Issue aims to provide a forum for researchers and practitioners to discuss and exchange recent advances, research results, and emerging research directions in Deep Learning Algorithm Generalization for Complex Industrial System, specifically in an environment of smart technologies such as artificial intelligence, big data, and so on. Hopefully, we can develop some new modeling and simulation methodologies to support the challenges of complex industrial systems, such as changes, adaptation, change, and innovativeness. This Special Issue covers a variety of contributions from different fields. 

Prof. Dr. Aboul Ella Hassanien
Prof. Dr. Ahmad Azar
Guest Editors

Manuscript Submission Information

Manuscripts should be submitted online at www.mdpi.com by registering and logging in to this website. Once you are registered, click here to go to the submission form. Manuscripts can be submitted until the deadline. All submissions that pass pre-check are peer-reviewed. Accepted papers will be published continuously in the journal (as soon as accepted) and will be listed together on the special issue website. Research articles, review articles as well as short communications are invited. For planned papers, a title and short abstract (about 250 words) can be sent to the Editorial Office for assessment.

Submitted manuscripts should not have been published previously, nor be under consideration for publication elsewhere (except conference proceedings papers). All manuscripts are thoroughly refereed through a single-blind peer-review process. A guide for authors and other relevant information for submission of manuscripts is available on the Instructions for Authors page. Electronics is an international peer-reviewed open access semimonthly journal published by MDPI.

Please visit the Instructions for Authors page before submitting a manuscript. The Article Processing Charge (APC) for publication in this open access journal is 2400 CHF (Swiss Francs). Submitted papers should be well formatted and use good English. Authors may use MDPI's English editing service prior to publication or during author revisions.