Research

Logistics operations of multi-national manufacturers are challenging in many perspectives. Large number of independent agents, products and dynamic retail demands complicate supply and inventory prediction. Optimization across independent operations is not possible and local optimization results in inefficient operations. Complexity not only increases the operational cost but also increases the failure rates and makes the fault recovery challenging.  We propose a scalable abstraction modeler for large logistics operations using large data analysis based on the information flow model. The abstraction modeler consists of three main components: large data monitor, logistics optimizer, and operation anomaly detector. Our key concepts are the information flow and the operation dependency for the characterization of large logistics applications. Using such characterization, we can accurately monitor and optimize large scale logistics operations. We can also detect, diagnose, and recover operation failures and performance degradation.

PIs: Hyong S. Kim (CMU), Akkarit Sangpetch (KMITL)

Fast and efficient bottle screening for recycling is an important problem in industrial plants. Recycling plants aim to recycle the good quality bottles by segregating them from bad quality bottles. The current system at TBR recycling plant uses a single threshold method to filter the bad bottles. However, this method resulted in high false positive rate. To address this issue, the in‐place screening system is replaced by manual filtering. But, this method reduced productivity rate. To solve the targeted problem of false negative rate and diminishing productivity rate, we came up with an idea to automate the entire bottle screening by the AI system. We propose to utilize the information about the texture and minor details of the bottle and build an AI classifier to derive to a filtering decision.

PIs: Marios Savvides (CMU), Akkarit Sangpetch (KMITL)

The goal of this research is to enable animal behavior monitoring through ground vibration collected with a set of low-cost Geophone sensors. Thailand’s livestock industry is a major driver of Thailand’s economy with roughly 6 million cattle (cows), 260 million chicken/ducks, 8 million pigs. Despite this number, this sector is expected to grow as export market to ASEAN countries is still low meat consumption. However, producers face many risks in producing and marketing of livestock. Weather may impair or kill livestock. Animal diseases may cause death or reduced productivity (mortality and morbidity). Feed supply, quality, and price may affect productivity and production expenses. Selling prices depend on the animal quality and uncertain global markets.

A key to mitigation of these risks is monitoring and detection of problems early before issues grow. In particular, any abnormal behavior needs to be detected early to ensure the quality and safety of the animals and workers.

PIs: Pei Zhang (CMU), Hae Young Noh (CMU), Orathai Sangpetch (KMITL)