Gianluca Nicchiotti
Gianluca Nicchiotti has a total of 4 co-authors. Here are the top co-authors by number of shared publications:
- Luc Fromaigeat
Meggitt Sensing Systems, CH-1701 Fribourg, Switzerland
2 shared publications - Fabrice Giuliani
Combustion Bay One e.U., Graz, Austria
1 shared publication - Krzysztof Soliński
Meggitt Sensing Systems, Rte de Moncor 4, 1701 Fribourg, Switzerland
1 shared publication - Lianel Etienne
École Polytechnique Fédérale de Lausanne, CH-1015 Lausanne, Switzerland
1 shared publication
Latest Publications
- Open Access
- 29 June 2021
- Published by PHM Society in Phm Society European Conference
- Vol. 6 (1) , 11
Lean Blowout Sensing and Processing via Optical Interferometry and Wavelet Analysis of Dynamic Pressure Data
This paper introduces a novel approach to monitor combustion instabilities in turbomachinery. This innovation is motivated by the need to better analyze flame health in a continuous flow machine, and more precisely in its hot core expressed by machine OEMs and end-users. This improved monitoring system will support an optimization of the machine performance, leading to a safer, cleaner, more flexible and more cost-efficient operation for the end-user. There have been several numerical studies that simulate the nonlinear dynamics of flame instabilities limit in simple combustors, they predict the pressure oscillations frequency is very low 5-25 Hz. Nowadays, in gas turbine monitoring, dynamic pressure measurement sensors are based on electrical working principle technologies such as piezoelectric, piezo-resistive and capacitive measurement. While exposed to high temperatures, due to pyro-electric effects and noise spectral density of charge amplifiers, piezo sensors signal to noise ratio (SNR) decreases sharply at low frequencies (about a factor 10 every 100°). Optical sensing systems provide much better stability, in the low frequency range when exposed to high temperature and, moreover, are intrinsically insensitive to vibration or EMI perturbation. This feature allows analyzing more accurately low frequency flame dynamics. Within the frame of a joint study between Meggitt, Combustion Bay One e.U. and FH Joanneum GmbH, the behavior of an optical measurement system was tested under different combustion instability conditions. By acting on system parameters (equivalent ratio, air speed) several tests have been conducted where combustion has been driven to different instability conditions such as premixed to diffusion flame transition, lean blow out, flashback. Data have been recorded and analysed from both the optical and the piezo sensors installed within the combustion chamber. The objective of the test campaign is twofold: compare optical to piezo system performances under flame combustion instabilities and, taking advantage of improved low frequency SNR of pressure data coming from optical sensors, to develop and validate an algorithmic strategy to detect/prognose phenomena such as Lean Blow Out and Flashback. The paper initially describes the principia of the used optical sensing technology, including some relevant hardware and software aspects of the measurement chain. After that, the experimental set-up and tests conditions are presented. About piezo to optical data comparison, SNR analysis of test data confirms the expectations of a lower noise at low frequency and the improved visibility of flame instabilities provided by optical sensor. Based on this, in order to analyze the transient behavior at low frequencies associated to flame instabilities two algorithmic strategies have been developed: wavelet analysis and chaos dimension analysis. Based on these methodologies several Lean Blow Out indicators have been implemented and tested. Details of their behavior in response to the variation of the equivalence ratio close to the flame blow out are presented and discussed. Three of the proposed indicators, under the investigated test conditions, proved to possess the capability to detect the blow out event. Depending on how the approach to the blow out event occurs, some indicators also shown prognostics capability. Results show some indicators start reacting as the flame approaches to extinction or as the equivalence ratio reduces. These indicators can be considered flame health meters. The test campaign results shown in the paper allow to conclude that optical sensing technologies provide advantages over piezoelectric sensing technologies such as: insensitivity to pyro-electric effects and inherent insensitivity to external perturbations, i.e. electromagnetic interferences and radio frequency interferences, vibrations. Particularly the improved accuracy at low frequencies of the sensor coupled with new signal processing techniques, as the ones presented, may support a flame health monitoring capability and the ability to sense blowout precursors. The replacement of piezo-electric with fiber optic sensors will therefore be desirable for many combustion monitoring applications, especially the ones requiring monitoring of low frequencies. This can, provide significant payoffs in engine reliability and operability, in enabling optimal performance over extended time periods as an engine ages, in reducing maintenance costs, and in increasing engine life.
- 29 June 2021
- Published by PHM Society in Phm Society European Conference
- Vol. 6 (1) , 11
Open AccessLean Blowout Sensing and Processing via Optical Interferometry and Wavelet Analysis of Dynamic Pressure Data
This paper introduces a novel approach to monitor combustion instabilities in turbomachinery. This innovation is motivated by the need to better analyze flame health in a continuous flow machine, and more precisely in its hot core expressed by machine OEMs and end-users. This improved monitoring system will support an optimization of the machine performance, leading to a safer, cleaner, more flexible and more cost-efficient operation for the end-user. There have been several numerical studies that simulate the nonlinear dynamics of flame instabilities limit in simple combustors, they predict the pressure oscillations frequency is very low 5-25 Hz. Nowadays, in gas turbine monitoring, dynamic pressure measurement sensors are based on electrical working principle technologies such as piezoelectric, piezo-resistive and capacitive measurement. While exposed to high temperatures, due to pyro-electric effects and noise spectral density of charge amplifiers, piezo sensors signal to noise ratio (SNR) decreases sharply at low frequencies (about a factor 10 every 100°). Optical sensing systems provide much better stability, in the low frequency range when exposed to high temperature and, moreover, are intrinsically insensitive to vibration or EMI perturbation. This feature allows analyzing more accurately low frequency flame dynamics. Within the frame of a joint study between Meggitt, Combustion Bay One e.U. and FH Joanneum GmbH, the behavior of an optical measurement system was tested under different combustion instability conditions. By acting on system parameters (equivalent ratio, air speed) several tests have been conducted where combustion has been driven to different instability conditions such as premixed to diffusion flame transition, lean blow out, flashback. Data have been recorded and analysed from both the optical and the piezo sensors installed within the combustion chamber. The objective of the test campaign is twofold: compare optical to piezo system performances under flame combustion instabilities and, taking advantage of improved low frequency SNR of pressure data coming from optical sensors, to develop and validate an algorithmic strategy to detect/prognose phenomena such as Lean Blow Out and Flashback. The paper initially describes the principia of the used optical sensing technology, including some relevant hardware and software aspects of the measurement chain. After that, the experimental set-up and tests conditions are presented. About piezo to optical data comparison, SNR analysis of test data confirms the expectations of a lower noise at low frequency and the improved visibility of flame instabilities provided by optical sensor. Based on this, in order to analyze the transient behavior at low frequencies associated to flame instabilities two algorithmic strategies have been developed: wavelet analysis and chaos dimension analysis. Based on these methodologies several Lean Blow Out indicators have been implemented and tested. Details of their behavior in response to the variation of the equivalence ratio close to the flame blow out are presented and discussed. Three of the proposed indicators, under the investigated test conditions, proved to possess the capability to detect the blow out event. Depending on how the approach to the blow out event occurs, some indicators also shown prognostics capability. Results show some indicators start reacting as the flame approaches to extinction or as the equivalence ratio reduces. These indicators can be considered flame health meters. The test campaign results shown in the paper allow to conclude that optical sensing technologies provide advantages over piezoelectric sensing technologies such as: insensitivity to pyro-electric effects and inherent insensitivity to external perturbations, i.e. electromagnetic interferences and radio frequency interferences, vibrations. Particularly the improved accuracy at low frequencies of the sensor coupled with new signal processing techniques, as the ones presented, may support a flame health monitoring capability and the ability to sense blowout precursors. The replacement of piezo-electric with fiber optic sensors will therefore be desirable for many combustion monitoring applications, especially the ones requiring monitoring of low frequencies. This can, provide significant payoffs in engine reliability and operability, in enabling optimal performance over extended time periods as an engine ages, in reducing maintenance costs, and in increasing engine life.
- Open Access
- 30 June 2018
- Published by PHM Society in Phm Society European Conference
- Vol. 4 (1)
Data-Driven Prediction of Unscheduled Maintenance Replacements in a Fleet of Commercial Aircrafts
Aerospace industries have become increasingly concerned about system availability and reliability. Data driven based prognosis is an emerging application aimed at building predictive models from readily available maintenance and operational databases. After validation, these models can be integrated into PHM systems to monitor equipment health and predict component failures before such events disrupt operations. For this project legacy data collected from two databases associated to a fleet of civil aircrafts during a period spanning 5 years have been used. The first database contains Central Management System (CMS) data (BIT messages and Flight Deck Effects), the second logs of maintenance activities. Part of the data collected from 2012 to mid 2015 have been used for the learning phase the rest spanning 2012-2016 period have been used for validation The goal is to predict failure events within an interval ranging from two to ten flights in advance to avoid unscheduled maintenance activities and operational disruptions. Hence two flights represent the minimal notice period/prognostics horizon whilst 10 flights is the maximal acceptable wasted life. Data-driven based prognostic uses pattern recognition and machine learning techniques to train historic data. In the proposed approach both techniques have been used. Through Support Vector Machine a prognostics anomaly detection step is initially performed to select the flight legs candidate for a prognostics alert. In a further step a subspace technique, borrowed from image processing domain and named Eigenface, allows to produce the signatures of the different types of maintenance actions and a template matching algorithm determines among the prognostics alert candidates the component to be replaced. Several tests have been conducted for different types of replacements and results will be presented using Receiver Operating Characteristic (ROC) curves and precision/recall metrics. Information contained in ROC allows the airliner to identify, according to its economic criteria, the optimal prognostics operating points.
- 30 June 2018
- Published by PHM Society in Phm Society European Conference
- Vol. 4 (1)
Open AccessData-Driven Prediction of Unscheduled Maintenance Replacements in a Fleet of Commercial Aircrafts
Aerospace industries have become increasingly concerned about system availability and reliability. Data driven based prognosis is an emerging application aimed at building predictive models from readily available maintenance and operational databases. After validation, these models can be integrated into PHM systems to monitor equipment health and predict component failures before such events disrupt operations. For this project legacy data collected from two databases associated to a fleet of civil aircrafts during a period spanning 5 years have been used. The first database contains Central Management System (CMS) data (BIT messages and Flight Deck Effects), the second logs of maintenance activities. Part of the data collected from 2012 to mid 2015 have been used for the learning phase the rest spanning 2012-2016 period have been used for validation The goal is to predict failure events within an interval ranging from two to ten flights in advance to avoid unscheduled maintenance activities and operational disruptions. Hence two flights represent the minimal notice period/prognostics horizon whilst 10 flights is the maximal acceptable wasted life. Data-driven based prognostic uses pattern recognition and machine learning techniques to train historic data. In the proposed approach both techniques have been used. Through Support Vector Machine a prognostics anomaly detection step is initially performed to select the flight legs candidate for a prognostics alert. In a further step a subspace technique, borrowed from image processing domain and named Eigenface, allows to produce the signatures of the different types of maintenance actions and a template matching algorithm determines among the prognostics alert candidates the component to be replaced. Several tests have been conducted for different types of replacements and results will be presented using Receiver Operating Characteristic (ROC) curves and precision/recall metrics. Information contained in ROC allows the airliner to identify, according to its economic criteria, the optimal prognostics operating points.
- Open Access
- 5 July 2016
- Published by PHM Society in Phm Society European Conference
- Vol. 3 (1)
Case Study: Vibration trip and post-event Analysis with Auto-Associative Neural Networks on a Large Steam Turbine
This 300 MW steam turbine at a coal-based thermal power plant is equipped with a protection system, a condition monitoring analysis software and an automatic diagnostic tool. The Machine Protection System (MPS) and Condition Monitoring System (CMS) configuration combines sensors, electronic hardware, firmware and software specific to this application. The protection system initiated a trip having identified high vibration. The trip prevented further damage. Subsequent analysis of the data using the condition monitoring software established the bearings most affected and pin pointed the source of high vibration. The data is post processed using an Auto-Associative Neural Networks (AANN) that has been trained with healthy data recorded several hours prior to the trip. AANN are methodologies widely used for novelty and anomaly detection. The AANN results indicates that such approach would be capable of detecting the failure event in advance compared to the automatic diagnostic system based on rules, demonstrating the validity of the approach in this context. Various aspects related to vibration: protection, condition monitoring, analysis, automatic diagnostics using rules and Neural Networks are presented and their results discussed.
- 5 July 2016
- Published by PHM Society in Phm Society European Conference
- Vol. 3 (1)
Open AccessCase Study: Vibration trip and post-event Analysis with Auto-Associative Neural Networks on a Large Steam Turbine
This 300 MW steam turbine at a coal-based thermal power plant is equipped with a protection system, a condition monitoring analysis software and an automatic diagnostic tool. The Machine Protection System (MPS) and Condition Monitoring System (CMS) configuration combines sensors, electronic hardware, firmware and software specific to this application. The protection system initiated a trip having identified high vibration. The trip prevented further damage. Subsequent analysis of the data using the condition monitoring software established the bearings most affected and pin pointed the source of high vibration. The data is post processed using an Auto-Associative Neural Networks (AANN) that has been trained with healthy data recorded several hours prior to the trip. AANN are methodologies widely used for novelty and anomaly detection. The AANN results indicates that such approach would be capable of detecting the failure event in advance compared to the automatic diagnostic system based on rules, demonstrating the validity of the approach in this context. Various aspects related to vibration: protection, condition monitoring, analysis, automatic diagnostics using rules and Neural Networks are presented and their results discussed.
- Open Access
- 5 July 2016
- Published by PHM Society in Phm Society European Conference
- Vol. 3 (1)
Machine Learning Strategy for Fault Classification Using Only Nominal Data
Machine learning methods are increasingly used for rotating machinery monitoring. Usually at set up, only data associated to an engine in a good state, the so called nominal data, are available for the machine learning phase. Nevertheless a classifier requires faulty data to be trained at identifying the causes of the anomalies and this fact has generally limited the usage of data driven approaches to fault detection tasks. The paper suggests a strategy to use machine learning methods even for fault classification purposes and diagnostics. Within the proposed framework three different machine learning methods, Gaussian Mixture Model (GMM), Support Vector Machines (SVM) and Auto Associative Neural Networks (AANN) have been implemented, tested and compared. The idea is to take into account some ‘a priori’ knowledge about the faults to be classified, to drive the behavior of the machine learning methodology (SVM or AANN or GMM) to be more or less reactive to the different faults. The indicators (features) more sensitive to each kind of fault are firstly selected on the basis of expert knowledge. For each different fault, a set of indicators is defined and computed from nominal data only. Each set is then used to produce training data for one specific fault. Such data sets are then used to train one instance of each method for each different fault. The underlying logic is that fault tuned input data is able to produce fault tuned instances of the methods. For example the instance trained with the indicators associated to a fault ‘A’ reacts more powerfully in presence of the fault ‘A’ than the others. Once an anomaly is detected, the comparison among the reactions of the different ‘fault tuned’ instances allows classifying the fault, not just to detect it. The results show best detection performances for SVM whilst AANN outperforms the other two methods for classification.
- 5 July 2016
- Published by PHM Society in Phm Society European Conference
- Vol. 3 (1)
Open AccessMachine Learning Strategy for Fault Classification Using Only Nominal Data
Machine learning methods are increasingly used for rotating machinery monitoring. Usually at set up, only data associated to an engine in a good state, the so called nominal data, are available for the machine learning phase. Nevertheless a classifier requires faulty data to be trained at identifying the causes of the anomalies and this fact has generally limited the usage of data driven approaches to fault detection tasks. The paper suggests a strategy to use machine learning methods even for fault classification purposes and diagnostics. Within the proposed framework three different machine learning methods, Gaussian Mixture Model (GMM), Support Vector Machines (SVM) and Auto Associative Neural Networks (AANN) have been implemented, tested and compared. The idea is to take into account some ‘a priori’ knowledge about the faults to be classified, to drive the behavior of the machine learning methodology (SVM or AANN or GMM) to be more or less reactive to the different faults. The indicators (features) more sensitive to each kind of fault are firstly selected on the basis of expert knowledge. For each different fault, a set of indicators is defined and computed from nominal data only. Each set is then used to produce training data for one specific fault. Such data sets are then used to train one instance of each method for each different fault. The underlying logic is that fault tuned input data is able to produce fault tuned instances of the methods. For example the instance trained with the indicators associated to a fault ‘A’ reacts more powerfully in presence of the fault ‘A’ than the others. Once an anomaly is detected, the comparison among the reactions of the different ‘fault tuned’ instances allows classifying the fault, not just to detect it. The results show best detection performances for SVM whilst AANN outperforms the other two methods for classification.