Sunil Menon

An algorithmic approach to adaptive state filtering using recurrent neural networks

Practical algorithms are presented for adaptive state filtering in nonlinear dynamic systems when the state equations are unknown. The state equations are constructively approximated using neural networks. The algorithms presented are based on the two-step prediction-update approach of the Kalman filter. The proposed algorithms make minimal assumptions regarding the underlying nonlinear dynamics and their noise statistics. Non-adaptive and adaptive state filtering algorithms are presented with both off-line and online learning stages. The algorithms are implemented using feedforward and recurrent neural network and comparisons are presented. Furthermore, extended Kalman filters (EKFs) are developed and compared to the filter algorithms proposed. For one of the case studies, the EKF converges but results in higher state estimation errors that the equivalent neural filters. For another, more complex case study with unknown system dynamics and noise statistics, the developed EKFs do not converge. The off-line trained neural state filters converge quite rapidly and exhibit acceptable performance. Online training further enhances the estimation accuracy of the developed adaptive filters, effectively decoupling the eventual filter accuracy from the accuracy of the process model.

Gain-scheduled nonlinear control of U-tube steam generator water level

Controller design for U-tube-steam generator (UTSG) water level at low operating powers is addressed via a systematic design procedure. The Linear Quadratic Gaussian with Loop Transfer Recovery met...

Fault detection and diagnosis in turbine engines using fuzzy logic

In this paper, we present a fuzzy logic based method of fault detection and diagnosis in gas turbine engines. The fuzzy logic system rule base is derived using heuristics extracted from designed experiments and flight data representing component performance changes due to field service degradation. The fuzzy logic rule based method incorporates both sensed engine parameters that represent non-deteriorated engine operation and fault conditions related to engine performance such as high pressure turbine, high pressure compressor and combustor deterioration. The fuzzy logic system is evaluated using residuals calculated based on both empirical models as inputs. The efficacy of the fuzzy logic system in detecting and diagnosing engine faults is demonstrated using field test data. We also examine performance robustness in the presence of varying levels of sensor noise and measurement errors.

Fault diagnosis in gas turbine engines using fuzzy logic

This paper describes a fuzzy logic-based method of fault detection and diagnosis in gas turbine engines. The fuzzy logic rule base is derived using heuristics based on designed experiments and flight data. The method is evaluated using model-based residuals and calculated values as inputs. The efficacy of the system is demonstrated using flight data. This paper describes how to augment a limited number of input parameters by combining them with the rates of change of the normal input parameters and other derived parameters. This augmented parameter set enables a better estimate of the prediction horizon for diagnosis. The paper also presents a case study where high-pressure spool deterioration is detected about two months prior to engine failure. Although, the system is demonstrated using the example of high pressure spool deterioration it can be applied to engine failures with similar characteristics.

Incipient Fault Detection and Diagnosis in Turbine Engines Using Hidden Markov Models

Incipient fault detection and diagnosis in turbine engines is key to effective maintenance and improved availability of systems dependent on these engines. In this paper, we present a novel method for incipient fault detection and diagnosis using Hidden Markov Models (HMMs). In particular, we focus on engine faults that are manifest in transient operating conditions such as engine startup and acceleration. HMMs are stochastic signal models that are effective in modeling transient signals. They are developed with engine data collected under nominal operating conditions. Engine data representing different fault conditions are used to develop the fault HMMs; a separate model is developed for each of the faults. Once the nominal and fault HMMs are developed, new engine data collected from the engine are evaluated against the HMMs and a determination is made whether a fault is indicated. Here, we demonstrate our HMM-based fault detection and diagnosis approach on engine speed profiles taken from a real engine. Further, the effectiveness of the HMM-based approach is compared with a neural-network-based approach and a method based on using principal component analysis in conjunction with a neural network approach.Copyright

Neural Network Models for Usage Based Remaining Life Computation

In engine structural life computations, it is common practice to assign a life of certain number of start-stop cycles based on a standard flight or mission. This is done during design through detailed calculations of stresses and temperatures for a standard flight, and the use of material property and failure models. The limitation of the design phase stress and temperature calculations is that they cannot take into account actual operating temperatures and stresses. This limitation results in either very conservative life estimates and subsequent wastage of good components or in catastrophic damage because of highly aggressive operational conditions, which were not accounted for in design. In order to improve significantly the accuracy of the life prediction, the component temperatures and stresses need to be computed for actual operating conditions. However, thermal and stress models are very detailed and complex, and it could take on the order of a few hours to complete a stress and temperature simulation of critical components for a flight. The objective of this work is to develop dynamic neural network models that would enable us to compute the stresses and temperatures at critical locations, in orders of magnitude less computation time than required by more detailed thermal and stress models. The current paper describes the development of a neural network model and the temperature results achieved in comparison with the original models for Honeywell turbine and compressor components. Given certain inputs such as engine speed and gas temperatures for the flight, the models compute the component critical location temperatures for the same flight in a very small fraction of time it would take the original thermal model to compute.

Adaptive state estimation using dynamic recurrent neural networks

The estimation of states from input and output measurements using linear state-space models has been widely studied. In particular, the Kalman filtering algorithm has found many applications, such as time-series forecasting, control, parameter estimation, and fault diagnosis. In this paper we propose a new method for adaptive state estimation using feedforward and recurrent neural networks and, in particular, for state filtering that is applicable to general nonlinear systems. The developed method has been applied to the problem of estimating the parameters of an electromechanical system consisting of a DC motor and a centrifugal pump with the associated pumping system. Finally, the proposed algorithm is used in estimating the states and a critical parameter of a complex process system, namely a heat exchanger.

An Adaptive State Filtering Algorithm for Systems With Partially Known Dynamics

On-line filtering of stochastic variables that are difficult or expensive to directly measure has been widely studied. In this paper a practical algorithm is presented for adaptive state filtering when the underlying nonlinear state equations are partially known. The unknown dynamics are constructively approximated using neural networks. The proposed algorithm is based on the two-step prediction-update approach of the Kalman Filter. The algorithm accounts for the unmodeled nonlinear dynamics and makes no assumptions regarding the system noise statistics. The proposed filter is implemented using static and dynamic feedforward neural networks. Both off-line and on-line learning algorithms are presented for training the filter networks. Two case studies are considered and comparisons with Extended Kalman Filters (EKFs) performed. For one of the case studies, the EKF converges but it results in higher state estimation errors than the equivalent neural filter with on-line learning. For another, more complex case study, the developed EKF does not converge. For both case studies, the off-line trained neural state filters converge quite rapidly and exhibit acceptable performance. On-line training further enhances filter performance, decoupling the eventual filter accuracy from the accuracy of the assumed system model. @DOI: 10.1115/1.1485747#

Startup fault detection and diagnosis in turbine engines

ln this paper, we present a novel method for startup fault detection and diagnosis in turbine engines. Traditional turbine engine fault detection and diagnosis methods are based on engine data collected at steady-state conditions. However, incipient faults are difficult to diagnose using steady-state engine data: only engine faults that are fairly developed can be detected using conventional methods. Because incipient engine component faults are often manifest in the engine startup characteristics, we present a method of characterizing the engine transient startup using the following steps: A feature vector is extracted from the measured engine sensor data during the engine startup. Then, several important discriminating features are distilled from the feature vector using principal component analysis (PCAj. The features obtained from this step are then classified using neural-network-based methods. The “leave-one-out” approach to cross-validation is applied to obtain an objective evaluation of the neural network training. The proposed fault detection and diagnosis method is evaluated using actual engine startup data and the results are presented.

A neural network-based speed filter for induction motors: Adapting to motor load changes

Effective sensorless speed estimation is desirable for both online condition monitoring of induction motor and sensorless adjustable speed AC drive applications. In this paper we present a neural network-based sensorless adaptive speed filter for induction motors. Only nameplate information and the actual motor currents and voltages are required for the initial setup of the proposed neural network-based speed filter. The speed filter gives acceptable steady state and transient speed response. The paper demonstrates the feasibility of adaptive speed filtering for induction motor which could be used for both diagnosis and control purposes.