Big Data Analytic Based on Scalable PANFIS for RFID Localization

Choiru Za'in, Mahardhika Pratama, Andri Ashfahani, Eric Pardede, Sheng Huang

Abstract

In modern industrial environments, real-time localization of manufacturing objects (MOs) using Radio Frequency Identification (RFID) is hindered by the high volume, velocity, and non-stationary nature of signal data. This paper introduces Scalable PANFIS, a distributed evolving neuro-fuzzy framework that leverages Apache Spark for parallel processing. By employing a novel rule-merging strategy, the system achieves a 20-fold increase in processing speed compared to traditional methods while maintaining high classification accuracy. Experimental results from a smart rack setup demonstrate the framework's effectiveness in handling massive RFID data streams with minimal accuracy loss.

Introduction

The advent of "smart" factories necessitates precise tracking of tools, personnel, and inventory. While RFID technology is cost-effective and widely deployed, analyzing Received Signal Strength (RSS) data presents significant challenges. RSS streams are characterized by high volume and velocity, non-stationarity due to environmental shifts (e.g., human movement or metallic interference), and high noise levels. Traditional machine learning models often require slow batch retraining, making them unsuitable for real-time applications. While "evolving" algorithms can learn incrementally, they typically lack the scalability required for big data environments.

Methodology

The proposed Scalable PANFIS framework addresses these challenges by adapting the Parsimonious Neo-Fuzzy Inference System (PANFIS) for a distributed architecture.

Distributed Processing with SparkR

To handle large-scale data, the system is integrated with Apache Spark. The data stream is partitioned into chunks, and each worker node in the cluster executes a local instance of PANFIS. This allows for parallelized learning across multiple servers, significantly reducing the computational bottleneck of single-machine systems.

Distributed architecture of Scalable PANFIS on Spark.

Rule Merging Scheme

A primary challenge in distributed learning is the redundancy of rules generated by independent nodes. Scalable PANFIS incorporates a Rule Merging Scheme to consolidate local models into a single, global model. This process utilizes the Bhattacharyya distance to identify overlapping fuzzy rules and performs homogeneity checks to ensure merging does not compromise precision. This approach ensures the final model remains parsimonious and interpretable without sacrificing accuracy.

The rule merging process for consolidating local models.

Experimental Evaluation

The framework was validated using a real-world dataset from the SIMTech laboratory in Singapore. The setup involved a "smart rack" where passive RFID tags were monitored by a reader. A dataset of 283,100 observations was collected over 20 hours, with the task of classifying objects into one of four shelf locations based on RSS frequency.

Performance Results

Scalable PANFIS demonstrated exceptional efficiency, processing the dataset in 104 seconds—a 20x speedup over the single-machine version (2,130 seconds). Despite the distributed nature of the processing, accuracy remained high at 96.67%, only a marginal 2% decrease from the centralized version (98.71%). Furthermore, the system significantly outperformed standard Spark MLlib algorithms, such as Generalized Linear Models (50.03%) and Gradient Boosted Trees (75.07%).

Conclusion

Scalable PANFIS provides a robust solution for real-time RFID localization in dynamic manufacturing settings. By balancing the trade-off between speed and accuracy through distributed processing and intelligent rule merging, the framework offers a scalable path for big data analytics. Future work will explore the application of this framework to higher-dimensional sensor fusion tasks and broader big data classification challenges.

Summary updated

Abstract

RFID technology has gained popularity to address localization problem in the manufacturing shopfloor by tracking the manufacturing object location to increase the production's efficiency. However, the signals (data) used for localization task is not easy to analyze because it is generated from the nonstationary environment. It also continuously arrive over time and yields the large-volume of data. Therefore, an advanced big data analytic is required to overcome this problem. We propose a distributed big data analytic framework based on PANFIS (Scalable PANFIS), where PANFIS is an evolving algorithm which has capability to learn data stream in the single pass mode. Scalable PANFIS can learn big data stream by processing many chunks/partitions of data stream. Scalable PANFIS is also equipped with rule' structure merging to eliminate the redundancy among rules. Scalable PANFIS is validated by measuring its performance against single PANFIS and other Spark's scalable machine learning algorithms. The result shows that Scalable PANFIS performs running time more than 20 times faster than single PANFIS. The rule merging process in Scalable PANFIS shows that there is no significant reduction of accuracy in classification task with 96.67 percent of accuracy in comparison with single PANFIS of 98.71 percent. Scalable PANFIS also generally outperforms some Spark MLib machine learnings to classify RFID data with the comparable speed in running time.
BibTeX
@inproceedings{zain2018big,
  author = {Choiru Za'in and Mahardhika Pratama and Andri Ashfahani and Eric Pardede and Sheng Huang},
  title = {{Big Data Analytic Based on Scalable PANFIS for RFID Localization}},
  booktitle = {SMC},
  year = {2018},
  url = {https://openreview.net/forum?id=yGFqW4faiO},
}
Figure