Andrew R. Mills

Bayesian framework for aerospace gas turbine engine prognostics

Prognostics is an emerging capability of modern health monitoring that aims to increase the fidelity of failure predictions. In the aerospace industry, it is a key technology to maximise aircraft availability, offering a route to increase time in-service and reduce operational disruption through improved asset management.

PWM Ripple Currents Based Turn Fault Detection for Multiphase Permanent Magnet Machines

Most permanent magnet (PM) machines are driven by inverters with pulse-width modulation (PWM) voltages. The currents contain high-frequency (HF) components which are inversely proportional to machine inductance. The HF PWM ripple currents can be used to detect a turn fault that gives rise to changes in inductance. The features of these HF components under turn fault conditions are analyzed. A bandpass filter is designed to extract the selected sideband components, and their root-mean-square (rms) values are measured. The rms values in all phases are compared. It is shown that the rms ripple current ratios between two adjacent phases provide a very good means of detecting turn fault with high signal-to-noise ratio. The detection method can identify the faulted phase and tolerate inherent imbalance of the machine, and is hardly affected by transient states. The method is assessed by simulations and experiments on a five-phase PM machine.

Heterogeneous hardware technologies for accelerating complex aerospace system simulations

This paper investigates the use of novel hardware and techniques to increase the speed of simulation for large gas turbine engine models. In particular, the work shows the results of attempting to accelerate an engine system model using multiple processor cores and a FPGA co-processor. Strengths and weaknesses of the technologies are illustrated and an account of the lessons learnt for distributing models over disparate technologies is provided. 1,2

Integrated equipment health management system design and development

Complex arguments need to be made regarding the benefit case for the sensor and algorithms available for health monitoring systems. A methodology to support these discussions is introduced based on functional failure information for the system. For the failures captured, incipient faults and their identifying effects are recorded. A corresponding algorithm and sensor input cost can be compared to the operational effects of not providing early warning of a failure. An intuitive visual heuristic, based upon an isolation distance metric and dimensional reduction technique, is introduced. The metric guides system architecture and evaluates the effectiveness of system inputs. A testing procedure, which is directly linked to unit and operational costs, demonstrates the heuristics validity against an on-wing expert system. The techniques are used to create a cost trade-off space, to compare health monitoring architectures. 1,2

Real Time Simulation in Floating Point Precision Using FPGA Computing

Real time simulations are indispensable for evaluation of new components, control system development, and system integration. There is a trade-off between model fidelity and the computational demands of the model; often lower fidelity models are chosen to speed development and to enable real time testing. FPGA technology offers an advantage over software simulation by exploiting bit and instruction level parallelism by default, but traditionally at the expense of coding effort and the need for experienced hardware engineers. The high cost of developing a model to execute on a FPGA is particularly prohibitive due to modifications occurring during system development. The work presents a process enabling an engineer to avoid hand-coded VHDL programming, yet take full advantage of the technology. The process described comes as a complete package: creating the model using pre-defined libraries, compilation and execution using specialized FPGA tools, simulation, co-simulation and interfacing with other technologies using in-house developed drivers.

Fusing an ensemble of diverse prognostic life predictions

Prognosis refers to the production of bounded estimates for the remaining useful life (RUL) of a component or system. Robust prognosis within well defined uncertainty bounds promises the ability to lower system sustainment costs and improve maintenance decision making. In ideal scenarios, the degradation of many assets is predictable from measurements taken during operation. These measurements, accompanied by appropriate knowledge, have been successfully used with various prognostic trending and prediction techniques to estimate the remaining useful life. However, there are scenarios where changes in operating environment or emerging fault modes can alter the degradation path an asset takes and thus invalidate previous predictions. One approach to overcoming these challenges is to develop prognostic techniques sensitive to change, however this sensitivity must be traded for robustness to imperfect degradation measurements. As an alternative, we develop multiple predictors, each with differing properties so that multiple estimates can be made based on different possible scenarios. A data fusion architecture is presented that combines multiple prognostic techniques weighted by their individual prediction uncertainty for the remaining useful life of an asset. The architecture is based on the concept of an ensemble Kalman Filter with an adaptive error co-variance parameter. The parameter is updated in real-time with the prediction variances produced by the set of prognostic predictions. Using real civil aerospace data, the technique is demonstrated to provide an amalgamated estimate of RUL to a fleet management operator. The developed architecture provides improved decision support to fleet maintainers in comparison to existing single prognostic technique methods.

Real World System Architecture Design Using Multi-criteria Optimization: A Case Study

System architecture design using multi-criteria optimization is demonstrated using a case study of an aero engine health management (EHM) system. A design process for optimal deployment of EHM system functional operations over physical architecture component locations, e.g., on-engine, on-aircraft and on-ground, is described. The EHM system architecture design needs to be optimized with respect to many qualitative criteria in terms of operational attributes within the constraints of resource limitations. In this paper the system architecture design problem is formulated as a multi-criteria optimization problem. Considering the large discrete search space of decision variables and many-objective functions and constraints, an evolutionary multi-objective genetic algorithm along with a progressive preference articulation technique, is used for solving the optimization problem. The optimization algorithm found a family of Pareto solutions which provided valuable insight into design trade-offs. Using the progressive preference articulation technique, the optimization search can be focused for the industrial decision maker on to a region of interest in the objective space. Performance of the proposed method is evaluated using various test metrics. Using this approach it was possible to identify the most significant design constraints (“hot spots”) and the opportunities afforded by either the relaxation or the tightening of these constraints, along with their performance implications.

Formal verification of a gain scheduling control scheme

Gain scheduling is a commonly used closed-loop control approach for safety critical non-linear systems, such as commercial gas turbine engines. It is preferred over more advanced control strategies due to a known route to certification. Nonetheless, the stability of the system is hard to prove analytically, and consequently, safety and airworthiness is achieved by burdensome extensive testing. Model checking can aid in bringing down development costs of such a control system and simultaneously improve safety by providing guarantees on properties of embedded control systems. Due to model-checking exhaustive verification capabilities, it has long been recognised that coverage and error-detection rate can be increased compared to traditional testing methods. However, the statespace explosion is still a major computational limitation when applying model-checking to verify dynamic system behaviour. A practical methodology to incrementally design and formally verify control system requirements for a gain scheduling scheme is demonstrated in this paper, overcoming the computational constraints traditionally imposed by model checking. In this manner, the gain-scheduled controller can be efficiently and safely generated with the aid of the model checker.

Turn fault detection for surface-mounted permanent magnet synchronous machine based on current residual

The mathematical models of the surface-mounted permanent magnet synchronous machine (SPMSM) in both healthy and turn fault conditions are utilized to generate the current residual with the same voltage input. The features in the current residual are analyzed with the consideration of model accuracy. The 2nd harmonic in the positive rotating dq reference or the dc component in the negative rotating dq reference can be extracted as the fault indicator. To estimate the dc component in the negative rotating reference that is more accurate and robust to transient states, a periodical integral is used in favor of conventional filtering or fast Fourier transform (FFT). It can be shown that the proposed fault detection scheme can distinguish the effect of speed or torque transient from the fault, thus giving a more reliable detecting result. The proposed technique is verified by simulation on a 12-slot, 14-pole SPMSM under various operating conditions.

PWM ripple currents based turn fault detection for 3-phase permanent magnet machines

Most permanent magnet (PM) machines are driven by inverter with pulse-width-modulation (PWM) voltages. The resultant phase current contains high frequency (HF) components, based on which, a new technique to detect turn fault is proposed. The machine model at high frequency is established and the HF current components are analyzed. A bandpass filter is used to extract the selected sideband components, and their root-mean-square (RMS) values are measured. The ratio of the RMS values in two adjacent phases are calculated. Both the individual ratio and the variance of 3 ratios are studied and testified to show the feasibility of detecting fault. Simulations have shown that both of them are valid fault indicator in ideal conditions while the variance is better in conditions with phase imbalance.