Cheol-Hong Min

This paper explores the integration of Large Language Models (LLMs) and secure Gen-AI technologies within engineering design and manufacturing, with a focus on improving inventory management, component selection, and recommendation workflows. The system is intended for deployment and evaluation in a real-world industrial environment. It utilizes vector embeddings, vector databases, and Approximate Nearest Neighbor (ANN) search algorithms to implement Retrieval-Augmented Generation (RAG), enabling context-aware searches for inventory items and addressing the limitations of traditional text-based methods. Built on an LLM framework enhanced by RAG, the system performs similarity-based retrieval and part recommendations while preserving data privacy through selective obfuscation using the ROT13 algorithm. In collaboration with an industry sponsor, real-world testing demonstrated strong results: 88.4% for Answer Relevance, 92.1% for Faithfulness, 80.2% for Context Recall, and 83.1% for Context Precision. These results demonstrate the system’s ability to deliver accurate and relevant responses while retrieving meaningful context and minimizing irrelevant information. Overall, the approach presents a practical and privacy-aware solution for manufacturing, bridging the gap between traditional inventory tools and modern AI capabilities and enabling more intelligent workflows in design and production processes.

Gas detection and environmental monitoring have become essential tasks to ensure the safety of people in many industries, such as mining, wastewater treatment plants, semiconductor manufacturing, and chemical manufacturers. To limit risk and keep people safe, it is necessary to monitor the environment where these gases could possibly be present and signal an alarm when a toxic or explosive condition may occur. This paper describes the implementation of a Low Power Wide Area Network (LPWAN) and the use of LoRa technology to design a cloud-based environmental and gas detection system. In the article, ESP32 microcontroller with integrated LoRa modules connected via SPI communication are used to wirelessly send various sensor readings back to the main controller, where dangerous situations will be announced. The various sensor readings include gas value, temperature, humidity, and some important diagnostic information from the end device to signal improper working conditions. Some of the sensors used in this project include electrochemical (gas) and capacitive (temperature and humidity) sensing elements. Since safety is the most important factor in these situations, if an unsafe condition is found, the device will set off an alarm immediately to communicate a problem for a safe evacuation protocol. Using LoRa technology, the data can be sent over large distances of over 1 kilometer to cover entire buildings with only one gateway/main controller, and low power consumption will require minimal maintenance and updates. The novel solution presented also offers real-time monitoring and predictive capabilities through cloud-enabled features and machine learning for a wider impact. Extensive testing and analysis of latency, power consumption, communication range, and reliability are presented along with practical guidelines.

Artificial intelligence has succeeded in many different areas in recent years. Especially the use of machine learning algorithms has been very popular in all areas, including fault detection. This paper explores a case study of applying machine learning techniques and neural networks to detect ten different machinery fault conditions using publicly available data sets collected from a tachometer, two accelerometers, and a microphone. Ten different conditions were classified using machine learning algorithms. Fifty-eight different features are extracted from time and frequency by applying the Short-Time Fourier Transform to the data with the window size of 1000 samples with 50% overlap. The Support Vector Machine models provided fault classification with 99.8% accuracy using all fifty-eight features. The proposed study explores the dimensionality reduction of the extracted features. Fifty-eight features were ranked using the Decision Tree model to identify the essential features as the classifier predictors. Based on feature extraction and raking, eleven predictors were extracted leading to reduced training complexity, while achieving a high classification accuracy of 99.7% could be obtained in less than half of the training time.

Field programmable gate arrays (FPGAs) have become widely prevalent in recent years as a great alternative to application-specific integrated circuits (ASIC) and as a potentially cheap alternative to expensive graphics processing units (GPUs). Introduced as a prototyping solution for ASIC, FPGAs are now widely popular in applications such as artificial intelligence (AI) and machine learning (ML) models that require processing data rapidly. As a relatively low-cost option to GPUs, FPGAs have the advantage of being reprogrammed to be used in almost any data-driven application. In this work, we propose an easily scalable and cost-effective cluster-based co-processing system using FPGAs for ML and AI applications that is easily reconfigured to the requirements of each user application. The aim is to introduce a clustering system of FPGA boards to improve the efficiency of the training component of machine learning algorithms. Our proposed configuration provides an opportunity to utilise relatively inexpensive FPGA development boards to produce a cluster without expert knowledge in VHDL, Verilog, or the system designs related to FPGA development. Consisting of two parts - a computer-based host application to control the cluster and an FPGA cluster connected through a high-speed Ethernet switch, allows the users to customise and adapt the system without much effort. The methods proposed in this paper provide the ability to utilise any FPGA board with an Ethernet port to be used as a part of the cluster and unboundedly scaled. To demonstrate the effectiveness of the proposed work, a two-part experiment to demonstrate the flexibility and portability of the proposed work - a homogeneous and heterogeneous cluster, was conducted with results compared against a desktop computer and combinations of FPGAs in two clusters. Data sets ranging from 60,000 to 14 million, including stroke prediction and covid-19, were used in conducting the experiments. Results suggest that the proposed system in this work performs close to 70% faster than a traditional computer with similar accuracy rates.
[GRAPHICS]

This study explores the structure of liquid/gas coaxial jets under forced and unforced conditions. The forcing is in the form of a transverse acoustic resonance within the confined space where the mixing occurs. The studied flows are relevant to combustion instabilities which involve an interaction between acoustic waves and reactant mixing. A variety of local and global signal processing methods were applied to digital flow visualization data to identify spatial and temporal features. The unforced case is in particular chaotic and influenced by a broad range of spatial and temporal phenomena. Proper orthogonal decomposition (POD) was able to extract flapping and convecting features, and spectral content of these behaviors is presented. The forced case results in organized structures that emerge above the background turbulence, including harmonics of the forcing frequency and nonlinear interactions between specific frequencies. The dynamic mode decomposition (DMD) performs the best in the forced case, clearly isolating all of these features. Wavelet analysis showed that forcing tended to reorganize energy from longer to shorter time scales. Bicoherence analysis of the data showed that the forcing causes a much different energy exchange in the outer and inner shear layers. The outer-to-inner jet coupling during forced conditions appears to be limited to an axial extent of about one to three inner jet diameters downstream of the jet exit. The recirculation zone between the inner and outer jet, extending about one inner jet diameter downstream, appears to disrupt the influence of forcing on the inner jet.

In this study, we target to automatically detect stereotypical behavioral patterns (stereotypy) and self-injurious behaviors (SIB) of Autistic children which can lead to critical damages or wounds as they tend to repeatedly harm oneself. Our custom designed accelerometer based wearable sensors are placed at wrists, ankles and upper body to detect stereotypy and SIB. The analysis was done on four children diagnosed with ASD who showed repeated behaviors that involve part of the body such as flapping arms, body rocking and self-injurious behaviors such as punching their face, or hitting their legs. Our goal of detecting novel events relies on the fact that the limitation of training data and variability in the possible combination of signals and events also make it impossible to design a single algorithm to understand all events in natural setting. Therefore, a semi-supervised method to discover and track unknown events in a multidimensional sensor data rises as a very important topic in classification and detection problems. In this paper, we show how the Higher Order Statistics (HOS) features can be used to design dictionaries and to detect novel events in a multichannel time series data. We explain our methods to detect novel events in a multidimensional time series data and combine the proposed semi-supervised learning method to improve the adaptability of the system while maintaining comparable detection accuracy as the supervised method. We, compare our results to the supervised methods that we have previously developed and show that although semi-supervised method do not achieve better performance compared to supervised methods, it can efficiently find new events and anomalies in multidimensional time series data with similar performance of the supervised method. We show that our proposed method achieves recall rate of 93.3% compared to 94.1% for the supervised method studied earlier.

Autism is one of the five pervasive development disorders that may cause severe impairment to a child. Depending on the degree of the symptoms, autism may cause severe impairments in one's social life such as social interaction and communication with other individuals. They may also face challenges in learning, concentrating, sensation and interacting with their surroundings. According to the Center for Disease Control (CDC), 1 in 150 8-year old children in many areas in the United States were diagnosed with autism. It is also known from recent studies that with early diagnosis we can intervene earlier which allows better assistance and treatment. Therefore, it is critical to have an objective assessment tool to assist diagnosis and for management. We have developed an affordable, reliable system that provides evidence based tools for assessment of children with autism. This system can detect various repetitive behavioral patterns often seen in children with autism and enables long term monitoring of repetitive behaviors. Therefore, it can be used to assist doctors, therapists, caregivers and parents with diagnosis and treatment of children with autism. This system incorporates 2 different sensor platforms which include environmental and wearable sensors. The system consists of a 3-axis accelerometer, small microcontroller and a Bluetooth module to transmit data to a base station such as a PC for analysis. We have customized this wearable device to integrate these modules which can be worn by a child. The environmental sensor configuration is composed of a microphone which records the acoustic data of the subject within the room. Using this sensor system, we are able to achieve the necessary information for assessment and therapy in autism research. We have analyzed the 3-axis accelerometer and acoustic data with an intelligent machine learning algorithm. The algorithm extracts time-domain and frequency domain features from the accelerometer data and applies statistical learning techniques to detect repetitive behavioral patterns. For acoustic data, we used sparse signal representation techniques to detect repetitive patterns that indicate vocalization behaviors. We have achieved an average of 89% in classification accuracy for detecting behavioral patterns. Based on the real data collected from children with autism, we were able to detect and recognize four self-stimulatory behaviors of children with autism. In one instance in which a subject had a tantrum, using the correlation between the hand flapping ratio and vocalization intensity, we were able to predict this extreme behavior. Our study opens an application in which devices could be used in a classroom environment to predict extreme behaviors in order that the stress of children with autism could be diverted accordingly so that their actions would be more socially agreeable.

In this paper, a single chip HDTV decoder is presented which drastically cuts down the cost of the HDTV receiver. The cost reduction is achieved by integration of the system parser, down-converting video decoder, and multi-mode audio decoder. System designers can also benefit from the features of our decoder which requires only 4 Mbytes of SDRAM for decoding the HDTV signal and various system-level functions such as the IEEE 1394 bus interface, I super(2)C bus controller and on-chip clock recovery function

In this paper, we present a method to reduce the blocking effects in block based coding schemes. Based on our observations, the blocking effect appears mainly in the regions where motion is so fast or complex that its motion is not well compensated. These regions tend to be coarsely quantized in the traditional method using only the spatial activity measure. In order to apply fine quantization on these regions, we introduce a new activity measure, the slope activity. By using this measure together with the spatial activity measure, we can perform an adaptive quantization on each macroblock in a picture. Our approach shows better picture quality compared to the MPEG2 TM5. Particularly in fast or complex motion scenes, the blocking effect in the object boundaries are far more reduced than that of MPEG2 TM5, therefore generating a much smoother image for comfortable viewing. Also, our method shows a higher PSNR than that of the MPEG2 TM5