Robust design using multiobjective optimisation and artificial neural networks with application to a heat pump radial compressor

Although robustness is an important consideration to guarantee the performance of designs under deviation, systems are often engineered by evaluating their performance exclusively at nominal conditions. Robustness is sometimes evaluated a posteriori through a sensitivity analysis, which does not guarantee optimality in terms of robustness. This article introduces an automated design framework based on multiobjective optimisation to evaluate robustness as an additional competing objective. Robustness is computed as a sampled hypervolume of imposed geometrical and operational deviations from the nominal point. In order to address the high number of additional evaluations needed to compute robustness, artificial neutral networks are used to generate fast and accurate surrogates of high-fidelity models. The identification of their hyperparameters is formulated as an optimisation problem. In the frame of a case study, the developed methodology was applied to the design of a small-scale turbocompressor. Robustness was included as an objective to be maximised alongside nominal efficiency and mass-flow range between surge and choke. An experimentally validated 1D radial turbocompressor meanline model was used to generate the training data. The optimisation results suggest a clear competition between efficiency, range and robustness, while the use of neural networks led to a speed-up by four orders of magnitude compared to the 1D code.

Robust multi-objective optimization of gasifier and solid oxide fuel cell plant for electricity production using wood

Ayse Dilan Celebi, François Maréchal, Shivom Sharma

Biomass is an attractive renewable and stored energy that can be converted to transportation fuels, chemicals and electricity using bio-chemical and thermo-chemical conversion routes. Notably, biofuels have relatively lower greenhouse gas emissions compared to the fossil fuels. A biomass gasifier can convert lignocellulosic biomass such as wood into syngas, which can be used in Solid Oxide Fuel Cell (SOFC) to produce heat and electricity. SOFC has very good thermodynamic conversion efficiency for converting methane or hydrogen into electricity, and integration of SOFC with gasifier gives heat integration opportunities that allow one to design systems with electricity production efficiencies as high as 70%. Generally, process design and operational optimization problems have conflicting performance objectives, and Multi-Objective Optimization (MOO) methods are applied to quantify the trade-offs among the objectives and to obtain the optimal values of design and operating parameters. This study optimizes biomass gasifier and SOFC plant for annual profit and annualized capital cost, simultaneously. A Pareto front has been obtained by solving Moo problem, and then net flow method is used to identify some optimal solutions from the Pareto front for the implementation into next phase. The constructed composite curves, which notify maximum amount of possible heat recovery, and first law efficiency also indicate better performance of the integrated plant. Uncertainty of market and operating parameters has been added to the optimization problem, and robust MOO of the integrated plant has been performed, which retains less sensitive Pareto solutions during the optimization. Finally, Pareto solutions obtained via normal and robust MOO approaches are considered for uncertainty analysis, and Pareto solutions obtained via robust MOO found to be less sensitive. (C) 2017 Elsevier Ltd. All rights reserved.

2017

A Mono Surrogate for Multiobjective Optimization

Ilya Loshchilov

Most surrogate approaches to multi-objective optimization build a surrogate model for each objective. These surrogates can be used inside a classical Evolutionary Multiobjective Optimization Algorithm (EMOA) in lieu of the actual objectives, without modifying the underlying EMOA; or to filter out points that the models predict to be uninteresting. In contrast, the proposed approach aims at building a global surrogate model defined on the decision space and tightly characterizing the current Pareto set and the dominated region, in order to speed up the evolution progress toward the true Pareto set. This surrogate model is specified by combining a One-class Support Vector Machine (SVMs) to characterize the dominated points, and a Regression SVM to clamp the Pareto front on a single value. The resulting surrogate model is then used within state-of-the-art EMOAs to pre-screen the individuals generated by application of standard variation operators. Empirical validation on classical MOO benchmark problems shows a significant reduction of the number of evaluations of the actual objective functions.

2010

Robustness and invariance properties of image classifiers

Apostolos Modas

Deep neural networks have achieved impressive results in many image classification tasks. However, since their performance is usually measured in controlled settings, it is important to ensure that their decisions remain correct when deployed in noisy environments. In fact, deep networks are not robust to a large variety of semantic-preserving image modifications, even to imperceptible image changes -- known as adversarial perturbations -- that can arbitrarily flip the prediction of a classifier. The poor robustness of image classifiers to small data distribution shifts raises serious concerns regarding their trustworthiness. To build reliable machine learning models, we must design principled methods to analyze and understand the mechanisms that shape robustness and invariance. This is exactly the focus of this thesis.First, we study the problem of computing sparse adversarial perturbations, and exploit the geometry of the decision boundaries of image classifiers for computing sparse perturbations very fast. We evaluate the robustness of deep networks to sparse adversarial perturbations in high-dimensional datasets, and reveal a qualitative correlation between the location of the perturbed pixels and the semantic features of the images. Such correlation suggests a deep connection between adversarial examples and the data features that image classifiers learn.To better understand this connection, we provide a geometric framework that connects the distance of data samples to the decision boundary, with the features existing in the data. We show that deep classifiers have a strong inductive bias towards invariance to non-discriminative features, and that adversarial training exploits this property to confer robustness. We demonstrate that the invariances of robust classifiers are useful in data-scarce domains, while the improved understanding of the data influence on the inductive bias of deep networks can be exploited to design more robust classifiers. Finally, we focus on the challenging problem of generalization to unforeseen corruptions of the data, and we propose a novel data augmentation scheme that relies on simple families of max-entropy image transformations to confer robustness to common corruptions. We analyze our method and demonstrate the importance of the mixing strategy on synthesizing corrupted images, and we reveal the robustness-accuracy trade-offs arising in the context of common corruptions. The controllable nature of our method permits to easily adapt it to other tasks and achieve robustness to distribution shifts in data-scarce applications.Overall, our results contribute to the understanding of the fundamental mechanisms of deep image classifiers, and pave the way for building more reliable machine learning systems that can be deployed in real-world environments.