Data-Driven Fault Diagnosis Techniques: Non-Linear Directional Residual vs. Machine-Learning-Based Methods.

Linear dependence of variables is a commonly used assumption in most diagnostic systems for which many robust methodologies have been developed over the years. In case the system nonlinearities are relevant, fault diagnosis methods, relying on the assumption of linearity, might potentially provide unsatisfactory results in terms of false alarms and missed detections. In recent years, many authors have proposed machine learning (ML) techniques to improve fault diagnosis performance to mitigate this problem. Although very powerful, these techniques require faulty data samples that are representative of any fault scenario. Additionally, ML techniques suffer from issues related to overfitting and unpredictable performance in regions which are not fully explored in the training phase. This paper proposes a non-linear additive model to characterize the non-linear redundancy relationships among the system signals. Using the multivariate adaptive regression splines (MARS) algorithm, these relationships are identified directly from the data. Next, the non-linear redundancy relationships are linearized to derive a local time-dependent fault signature matrix. The faulty sensor can then be isolated by measuring the angular distance between the column vectors of the fault signature matrix and the primary residual vector. A quantitative analysis of fault isolation and fault estimation performance is performed by exploiting real data from multiple flights of a semi-autonomous aircraft, thus allowing a detailed quantitative comparison with state-of-the-art machine-learning-based fault diagnosis algorithms.

A Comprehensive Case Study of Data-Driven Methods for Robust Aircraft Sensor Fault Isolation.

Recent catastrophic events in aviation have shown that current fault diagnosis schemes may not be enough to ensure a reliable and prompt sensor fault diagnosis. This paper describes a comparative analysis of consolidated data-driven sensor Fault Isolation (FI) and Fault Estimation (FE) techniques using flight data. Linear regression models, identified from data, are derived to build primary and transformed residuals. These residuals are then implemented to develop fault isolation schemes for 14 sensors of a semi-autonomous aircraft. Specifically, directional Mahalanobis distance-based and fault reconstruction-based techniques are compared in terms of their FI and FE performance. Then, a bank of Bayesian filters is proposed to compute, in flight, the fault belief for each sensor. Both the training and the validation of the schemes are performed using data from multiple flights. Artificial faults are injected into the fault-free sensor measurements to reproduce the occurrence of failures. A detailed evaluation of the techniques in terms of FI and FE performance is presented for failures on the air-data sensors, with special emphasis on the True Air Speed (TAS), Angle of Attack (AoA), and Angle of Sideslip (AoS) sensors.

Influence of hormonal therapy in prostate cancer patients undergoing [18F]fluoromethyl choline PET/CT: a retrospective study.

BACKGROUND: Patients with suspected recurrence of prostate cancer undergoing [18F]fluoromethyl choline ([18F]FCH) PET/CT were retrospectively evaluated to investigate the influence of hormonal therapy (HT) in [18F]FCH uptake. METHODS: [18F]FCH PET/CT was performed in 102 surgically treated patients with suspected recurrence (PSA increase >0.2 ng/mL) of prostate cancer, divided in two groups: under HT (N.=54) and without HT (N.=48) at the time of PET scanning. PET/CT was carried out by an integrated system (Biograph 6, CTI/Siemens, Knoxville, TN, USA) intravenously by administering 4.1 MBq/kg of [18F]FCH to each patient; images were acquired 60 minutes later. RESULTS: On the total number of patients, 66 were found to be true positives (TP), 9 false positives (FP), 5 false negatives (FN) and 22 true negatives (TN), sensitivity to [18F]FCH PET/CT was 93%, specificity 71%, accuracy 86%, positive predictive value (PPV) 88%, negative predictive value (NPV) 81%. In the 54 patients under HT, 38 were TP, 6 FP, 3 FN and 7 TN, sensitivity was 93%, specificity 54%, accuracy 83%, PPV 86% and NPV was 70%. In the 48 patients receiving no HT, 28 were TP, 3 FP, 2 FN and 15 TN, sensitivity was 93%, specificity 83%, accuracy 90%, PPV 90% and NPV 88%. A χ2 test showed that sensitivity, accuracy and PPV did not differ among patients with and without HT, while specificity and NPV were significantly lower (P<0.001) in HT treated patients. CONCLUSIONS: Sensitivity, accuracy and PPV were similar in patients with and without HT. Specificity and NPV were reduced in patients under HT, but further data are necessary to support if this reduction is casual or related to therapy and it could be confirmed in a larger series of patients.

Design of a neural network adaptive controller via a constrained invariant ellipsoids technique.

In safety critical applications, control architectures based on adaptive neural networks (NNs) must satisfy strict design specifications. This paper presents a practical approach for designing a mixed linear/adaptive model reference controller that recovers the performance of a reference model, and guarantees the boundedness of the tracking error within an a priori specified compact domain, in the presence of bounded uncertainties. The linear part of the controller results from the solution of an optimization problem where specifications are expressed as linear matrix inequality constraints. The linear controller is then augmented with a general adaptive NN that compensates for the uncertainties. The only requirement for the NN is that its output must be confined within pre-specified saturation limits. Toward this end a specific NN output confinement algorithm is proposed in this paper. The main advantages of the proposed approach are that requirements in terms of worst-case performance can be easily defined during the design phase, and that the design of the adaptation mechanism is largely independent from the synthesis of the linear controller. A numerical example is used to illustrate the design methodology.