A Novel GAN-Based Production Data Imputation Framework for Smart Manufacturing

Manufacturing, serving as a pivotal dynamo propelling modern economic development, retains an indomitable, unrivaled preeminence. In recent epochs, bolstered by unceasing enhancements in the realms of information and digitization, intelligent manufacturing is catalyzing the trans Formation of the manufacturing sector from traditional paradigms to bespoke, high-caliber production. Intelligent manufacturing, functioning as the linchpin for the evolution and elevation of the manufacturing industry, is fundamentally underpinned by data. Nevertheless, due to pitfalls such as personnel oversight, network packet loss, and cybersecurity attacks, industrial big data is plagued by a significant issue of data omission. To address the pragmatic necessities of production data imputation within a personalized customization setting, and acknowledging the dynamic and recurrently fluctuating traits of industrial data, this paper introduces a production data imputation framework predicated on Conditional Generative Adversarial Networks(CGAN). Within this framework, we initially procure key indicative information through the binary encoding of production tasks. Subsequently, we devise a time-series forecaster and a relationship learning module to extract input information more congruent with the target distribution from the raw data. Furthermore, to enhance the training efficacy of the gen generator, We incorporated production data processed via random masking for ancillary training. Lastly, we executed comparative and ablation experiments on an authentic dataset. The experimental outcomes lucidly substantially ate the efficacy of this proposed framework.