Production flow modeling based on BLE-based RSSI data with non-detectable areas

Abstract

This study presents a method for modelling manufacturing processes to predict key performance indicators (KPIs) such as cycle time using Bluetooth Low Energy (BLE) data. We consider BLE applications to be similar to Radio-Frequency IDentification (RFID) scenarios, with a single BLE scanner indicating a single working area. This work considers the case when Received Signal Strength Indicator (RSSI) data are unavailable in some areas, such as, when products are in temporary storage areas away from the production areas. We solve this problem with a Duration and Interval Hidden Markov Model (DI-HMM), in which time spent in production areas is represented as duration and those with absence data as intervals. To parameterize the DIHMM model, we propose a two-stage machine-learning problem based on a classification tree and a Hidden Semi Markov Model (HSMM). To investigate the proposed model, the RSSI observation sequences are generated using MATLAB Bluetooth Toolbox and real-world experimentation. The runtime scenario compares estimated and original states, and the average accuracy of 100 test sequences is around 95%. In the offline forecast scenario, an estimated DI-HMM parameter is used to forecast 200 sequences, then compared with sequences with a vector distance with a similarity score of 0.4717.

Keywords

• manufacturing process
• bluetooth low energy
• duration and interval hidden markov model
• classification tree
• hidden semi markov model