A5069430909- Refrigeration and Air Conditioning Technologies
- Building Energy and Comfort Optimization
- Advanced Control Systems Optimization
- Heat Transfer and Optimization
- Model Reduction and Neural Networks
- Heat Transfer and Boiling Studies
- Fault Detection and Control Systems
- Advanced Multi-Objective Optimization Algorithms
- Extremum Seeking Control Systems
- Modeling and Simulation Systems
- Neural Networks and Applications
- Gaussian Processes and Bayesian Inference
- Gas Dynamics and Kinetic Theory
- Energy Load and Power Forecasting
- Wind and Air Flow Studies
- Power Quality and Harmonics
- Simulation Techniques and Applications
- Structural Health Monitoring Techniques
- Hydraulic and Pneumatic Systems
- Phase Equilibria and Thermodynamics
- Magnetic confinement fusion research
- Reservoir Engineering and Simulation Methods
- Fusion materials and technologies
- Control Systems and Identification
- Nuclear Engineering Thermal-Hydraulics
Mitsubishi Electric (United States)
2016-2025
Mitsubishi Electric (Japan)
2014
Massachusetts Institute of Technology
2000-2010
Research Institute for Electromagnetic Materials
2003
Nonintrusive load monitoring (NILM) can determine operating schedule of electrical loads in a target system from measurements made at centralized location, such as the electric utility service entry. NILM is an ideal platform for extracting useful information about any that uses electromechanical devices. It has low installation cost and high reliability because it bare minimum sensors. possible to use modem state parameter estimation algorithms verify remotely "health" by using analyze...
We present a gradient-based meta-learning framework for rapid adaptation of neural state-space models (NSSMs) black-box system identification. When applicable, we also incorporate domain-specific physical constraints to improve the accuracy NSSM. The major benefit our approach is that instead relying solely on data from single target system, utilizes diverse set source systems, enabling learning limited data, as well with few online training iterations. Through benchmark examples,...
This paper describes the integration of generative deep learning models for data-driven building energy simulation. The (GMs) are trained to learn distributions input signals from data using Python and PyTorch interfaced with physics-based Modelica models. developed requirements provide background on typical needs that focus simulation performance. Simulation examples Buildings library, refactored receive GM inputs, presented illustrate benefits proposed approach how GMs can be used...
This paper presents two methods for reallizing fluid property functions in Modelica simulation models. Each makes use of a coordinate transformation that aligns one with the saturation curve. provides precise representation function at curve, and connected domains interest including liquid, vapor, supercritical two-phase regions. Both approaches make spline approximation aligned coordinates, are numerically efficient, well conditioned, allow efficient calculation derivatives up to any...
Nonintrusive load monitoring (NILM) is accomplished by sampling voltage and current at high rates reducing the resulting start transients or harmonic contents to concise "signatures." Changes in these signatures can be used detect, many cases directly diagnose, equipment component faults associated with rooftop cooling units. Use of NILM for fault detection diagnosis (FDD) important because (1) it complements other FDD schemes that are based on thermo-fluid sensors analyses (2) minimally...
This paper presents a Kalman-filter approach for computing spectral envelopes of current waveforms nonintrusive load monitoring on the electric utility. Spectral represent time-varying frequency content and phase relative to voltage. Thus, techniques presented in this may be applicable variety lock-in measurement signal processing techniques. The computational performance proposed method favorably compares previous efforts. is demonstrated with data from field.
Deep neural state-space models (SSMs) provide a powerful tool for modeling dynamical systems solely using operational data. Typically, SSMs are trained data collected from the actual system under consideration, despite likely existence of similar which have previously been deployed in field. In this paper, we propose use model-agnostic meta-learning (MAML) constructing deep encoder network-based SSMs, by leveraging combination archived (used to meta-train offline) and limited rapid online...
This paper describes two hardware prototypes and estimation schemes for determining the parameters of a simple, physically based, point-of-use electric utility model using transient measurements. Parameters are estimated data collected by prototypes. Frequency-dependent effects observed in previous work this area modeled. Performance techniques given is demonstrated comparison measured predicted line voltage distortion during current transients created several loads.
Field studies have demonstrated that the non-intrusive load monitor (NILM) can effectively evaluate state of many electromechanical systems by analyzing electrical power they draw. This paper discusses NILM applications in marine environment. Machinery data collected from USCGC SENECA (WMEC-906), a 270-foot U.S. Coast Guard cutter, indicates successfully diagnose failure flexible couplings and presence leaks cycling systems. both these shipboard problems, it details methodology used to...
We study the problem of performance optimization closed-loop control systems with unmodeled dynamics. Bayesian (BO) has been demonstrated effective for improving by automatically tuning controller gains or reference setpoints in a model-free manner. However, BO methods have rarely tested on dynamical constraints. In this paper, we propose violation-aware algorithm (VABO) that optimizes while simultaneously learning constraintfeasible solutions. Unlike classical constrained which allow an...
Harmonic analysis of motor current has been used to track the speed motors for sensorless control. Algorithms exist that a given dedicated stator measurement, example. also applied diagnostic detection electro-mechanical faults such as damaged bearings and rotor eccentricity. This paper demonstrates utility harmonic fault diagnostics in non-intrusive monitoring applications, where multiple loads are tracked by sensor only aggregate service. An optimization routine is implemented maintain...
This article compares the effects of two different refrigerant flow modeling assumptions on transient performance vapor-compression heat pump cycles. comparison is done a dynamic system-level model flash tank vapor injection cycle that includes finite-volume exchanger models. The effect and specific slip ratio correlations both equilibrium operating point behavior are demonstrated through simulations experiments. It shown equivalent with each have mass inventories some aspects system...
Advances in modeling and computation have resulted high-fidelity digital twins capable of simulating the dynamics a wide range industrial systems. These simulation models often require calibration, or estimation an optimal set parameters some goodness-of-fit sense, to reflect system's observed behavior. While searching over parameter space is inevitable part calibration process, are rarely designed be valid for arbitrarily large spaces. The application existing methods, therefore, results...