Lightweight Industrial Cohorted Federated Learning for Heterogeneous Assets

TL;DR

The paper addresses performance degradation in federated learning caused by heterogeneous industrial data arising from varying machine types, firmware, and operating conditions. It introduces Lightweight Industrial Cohorted FL (LICFL), which cohorts clients using only model parameters, enabling personalized training without added on-edge computation or communication, and extends it with Adaptive LICFL (ALICFL) that auto-selects aggregation strategies per cohort. LICFL leverages PCA and spectral clustering on flattened model parameters to form cohorts, while ALICFL employs momentum-based aggregations (e.g., FedAvg, FedAdam, FedAdaGrad, FedYogi) chosen per round for each cohort to speed convergence. Experimental evaluation on a predictive maintenance dataset with 100 machines shows LICFL and especially ALICFL achieve superior client-level and global performance, highlighting practical benefits for edge analytics in Industry 4.0 contexts.

Abstract

Federated Learning (FL) is the most widely adopted collaborative learning approach for training decentralized Machine Learning (ML) models by exchanging learning between clients without sharing the data and compromising privacy. However, since great data similarity or homogeneity is taken for granted in all FL tasks, FL is still not specifically designed for the industrial setting. Rarely this is the case in industrial data because there are differences in machine type, firmware version, operational conditions, environmental factors, and hence, data distribution. Albeit its popularity, it has been observed that FL performance degrades if the clients have heterogeneous data distributions. Therefore, we propose a Lightweight Industrial Cohorted FL (LICFL) algorithm that uses model parameters for cohorting without any additional on-edge (clientlevel) computations and communications than standard FL and mitigates the shortcomings from data heterogeneity in industrial applications. Our approach enhances client-level model performance by allowing them to collaborate with similar clients and train more specialized or personalized models. Also, we propose an adaptive aggregation algorithm that extends the LICFL to Adaptive LICFL (ALICFL) for further improving the global model performance and speeding up the convergence. Through numerical experiments on real-time data, we demonstrate the efficacy of the proposed algorithms and compare the performance with existing approaches.

Figures (9)

• Figure 1: Federated Learning, each client updates the model parameters with local data and sends them to the server, which aggregates and shares them back.
• Figure 2: Industrial Federated Learning, machine type is considered as a filter or parameter for cohorting to avoid divergence of the global model and improve performance at the client-level model.
• Figure 3: Functional architecture of Adaptive lightweight industrial cohorted federated learning ($\texttt{ALICFL}$) method.
• Figure 4: Effect of primary-level cohorting (using meta information) on client-level performance. Proposed algorithm $\texttt{LICFL}$ along with primary-level cohorting is labeled as $\texttt{LICFL}_{M}$. $\texttt{LICFL}_{M}$ outperforms the vanilla $\texttt{FL}$ and $\texttt{LICFL}$ at all considered clients.
• Figure 5: Comparison of global model performance at server against communication rounds. The proposed approach $\texttt{LICFL}$ provides a lower loss value and converges faster than baseline approaches.