This thesis focuses on Artificial Intelligence-driven design and on Noise, Vibration, and Harshness (NVH) analysis of automotive electric motors. The first objective of this research is to introduce Artificial Intelligence (AI) techniques to support and enhance the development of electric traction motors. The performance of the motor must be ensured both during the design phase and throughout its entire operational lifespan. Initially, AI methods were adopted to implement a multi-objective optimal design process for automotive electric motors. In particular, Artificial Neural Networks (ANNs) were employed to globally approximate the motor performance simulated by a Finite Element (FE) model. These surrogate models, combined with a newly developed optimization algorithm, the Adaptive Pareto Algorithm (APA), based on the iterative definition of the Pareto front, were used to optimize traction electric motors. The accuracy of the ANN was progressively improved through multiple infill operations guided by the APA, achieving a high level of predictive accuracy. The integration of Artificial Intelligence significantly increased the efficiency of the optimization process by reducing computational time, enabling the exploration of a broader design space and generating solutions that outperform those obtained through traditional design methods. Electric motor optimization was also performed within a robust design framework. AI techniques were further leveraged during the monitoring phase to detect faults in Permanent Magnet Synchronous Motors (PMSMs) under durability testing. Autoencoders (AE) were adopted as unsupervised learning techniques for this task, due to their ability to automatically extract features and patterns from operational data. After a comparative analysis of different Autoencoder architectures, a one-dimensional Autoencoder Convolutional Neural Network (1D AE-CNN) was selected as the most effective solution. Based on NVH data, the 1D AE-CNN was employed as a fault detection strategy for PMSMs operating under heavy load cycles characterized by high speed and torque variations. The implemented fault detection algorithm, tested on real motor operational data, successfully identified various PMSM faults, surpassing the limitations of traditional diagnostic methods. The AI-based fault detection system supports proactive maintenance, thereby reducing the risk of critical failures and enhancing overall system reliability. The second objective of this thesis is to study the NVH characteristics of PMSMs to develop a comprehensive understanding of their vibroacoustic behavior and guide the design of effective noise countermeasures. Electric traction motors play a major role in the overall noise emissions of electric vehicles. Moreover, they produce high-frequency tonal noise, which is particularly unpleasant for passengers. This highlights the need for a dedicated study to ensure both acoustic comfort and sound quality. To this end, an overview of current NVH challenges in the design of PMSMs for automotive applications has been carried out. Particular attention is devoted to the analytical and numerical methods used in NVH modeling, which consists of three main steps: electromagnetic force computation, motor structural modeling and acoustic emission analysis. Special focus is also placed on experimental techniques for the vibroacoustic characterization of electric motors through modal analysis. In addition to the well-established Experimental Modal Analysis (EMA) method, vibration response measurements during motor run-ups are exploited to apply Operational Deflection Shapes (ODS), Operational Modal Analysis (OMA) and Order-Based Modal Analysis (OBMA). Notably, OBMA is applied for the first time to electric motors within the context of this work. The advantages and limitations of the different techniques are compared through the study of a PMSM. The complementary use of the presented methods provides a comprehensive modal characterization of the electric motor. Finally, following a fully numerical approach, several NVH predictive models of electric motors have been employed to develop effective countermeasure techniques aimed at reducing noise emissions. The proposed solutions focus on late-design interventions that do not require modifications to the motor geometry. The first approach is based on the application of current injection strategies. This study presents an experimentally validated numerical simulation framework capable of predicting the noise reduction achievable through such strategies. The entire process is modeled, from current injection to final acoustic emission, across all simulation steps, enabling the formulation of an optimization problem for noise minimization. Using this method, a noise reduction of up to 27 dB has been achieved at the resonant frequencies of the motor. A second novel countermeasure technique to improve the NVH performance of electric machines involves applying a constrained viscoelastic rubber layer on the outer surface of a PMSM as a vibration damping treatment. The study assesses the effectiveness of this countermeasure in reducing acoustic emissions at different temperatures through a combination of numerical modeling and experimental validation. The frequency-dependent behavior of the material is accounted for using a complex modulus approach. The application of a viscoelastic layer with a thickness of 1.25 mm leads to a maximum noise reduction of 9 dB.
Questa tesi si concentra sulla progettazione guidata dall’Intelligenza Artificiale e sull’analisi NVH (Noise, Vibration, and Harshness) dei motori elettrici per applicazioni automobilistiche. Il primo obiettivo di questa ricerca è introdurre tecniche di Intelligenza Artificiale (IA) a supporto dello sviluppo dei motori elettrici di trazione. Le prestazioni del motore devono essere garantite sia nella fase di progettazione sia durante l’intero ciclo di vita del componente. Inizialmente, metodi di IA sono stati adottati per implementare un processo di progettazione ottimale multi-obiettivo per motori elettrici. In particolare, le Artificial Neural Networks (ANNs) sono state utilizzate per approssimare le prestazioni del motore simulate tramite un modello agli Elementi Finiti (FE). Questi modelli surrogati, combinati con un nuovo algoritmo di ottimizzazione, l’Adaptive Pareto Algorithm (APA), basato sulla definizione iterativa del fronte di Pareto, sono stati impiegati per ottimizzare i motori elettrici di trazione. L’accuratezza delle ANN è stata progressivamente migliorata attraverso operazioni di infittimento guidate dall’APA, raggiungendo un elevato livello di precisione predittiva. L’integrazione dell’Intelligenza Artificiale ha aumentato significativamente l’efficienza del processo di ottimizzazione, riducendo i tempi di calcolo, permettendo l’esplorazione di uno spazio delle variabili più ampio e generando soluzioni migliori rispetto a quelle ottenute con i metodi tradizionali. L’ottimizzazione del motore elettrico è stata inoltre condotta in un contesto di progettazione robusta. Le tecniche di IA sono state impiegate anche nella fase di monitoraggio, durante test di durata, per rilevare guasti nei motori sincroni a magneti permanenti (PMSMs). A questo scopo, gli Autoencoder (AE) sono stati adottati come tecniche di apprendimento non supervisionato, grazie alla loro capacità di estrarre automaticamente caratteristiche e pattern dai dati operativi. Dopo un’analisi comparativa di diverse architetture di rete, un AE composto da 1-Dimension Convolutional Neural Network (1D AE-CNN) è stato selezionato come soluzione più efficace. Basandosi su dati NVH, la 1D AE-CNN è stata utilizzata come strategia di rilevamento guasti per PMSMs operanti in cicli di durata caratterizzati da profili con elevate variazioni di velocità e coppia. L’algoritmo di fault detection implementato, testato su dati reali di funzionamento del motore, ha identificato con successo diversi guasti nei PMSMs, superando i limiti dei metodi diagnostici tradizionali. Il sistema di rilevamento guasti basato su IA supporta una manutenzione proattiva, riducendo il rischio di guasti critici e migliorando l’affidabilità complessiva del sistema. Il secondo obiettivo di questa tesi è studiare le caratteristiche NVH dei PMSMs al fine di sviluppare una conoscenza completa del loro comportamento vibroacustico e guidare quindi la progettazione di efficaci contromisure acustiche. I motori elettrici di trazione rappresentano una delle principali fonti di emissione sonora nei veicoli elettrici. Inoltre, essi producono suoni ad alta frequenza, particolarmente fastidiosi per i passeggeri del veicolo. Ciò evidenzia la necessità di uno studio dedicato per garantire sia il comfort acustico sia la qualità sonora. A tal fine, è stata condotta un’analisi delle attuali problematiche NVH nella progettazione di PMSMs per applicazioni automobilistiche. Particolare attenzione è dedicata ai metodi analitici e numerici utilizzati nella modellazione NVH, che si compone di tre fasi principali: calcolo delle forze elettromagnetiche, modellazione strutturale del motore e analisi delle emissioni acustiche. Un’attenzione particolare è rivolta anche alle tecniche sperimentali per la caratterizzazione vibroacustica dei motori elettrici tramite analisi modale. Oltre al metodo consolidato dell’Experimental Modal Analysis (EMA), le misure di vibrazioni durante le rampe in velocità del motore sono state utilizzate per applicare le tecniche di Operational Deflection Shapes (ODS), Operational Modal Analysis (OMA) e Order-Based Modal Analysis (OBMA). In particolare, l’OBMA viene applicata per la prima volta ai motori elettrici in questa ricerca. I vantaggi e i limiti delle diverse tecniche vengono confrontati attraverso lo studio di un PMSM. L’uso complementare dei metodi presentati fornisce una caratterizzazione modale completa del motore elettrico. Infine, adottando un approccio completamente numerico, sono stati impiegati diversi modelli predittivi NVH di motori elettrici per sviluppare tecniche di contromisura acustica efficaci, volte a ridurre le emissioni sonore. Le soluzioni proposte si concentrano su interventi in fase avanzata di progettazione, che non richiedono modifiche alla geometria del motore. Il primo approccio è basato sull’applicazione di strategie di iniezione di corrente. Questo studio presenta un framework di simulazione numerica validato sperimentalmente, in grado di prevedere la riduzione del rumore ottenibile tramite tali strategie. L’intero processo viene modellato, dall’iniezione di corrente all’emissione acustica finale, coprendo tutti i passaggi di simulazione, e consentendo la formulazione di un problema di ottimizzazione per la minimizzazione del rumore. Con questo metodo, è stata ottenuta una riduzione del rumore fino a 27 dB in corrispondenza delle frequenze di risonanza del motore. Una seconda tecnica innovativa di contromisura per migliorare le prestazioni NVH delle macchine elettriche prevede l’applicazione di uno strato di gomma viscoelastica vincolata sulla superficie esterna di un PMSM come trattamento di smorzamento delle vibrazioni. Lo studio valuta l’efficacia di questa contromisura nella riduzione delle emissioni acustiche a diverse temperature, attraverso una combinazione di modellazione numerica e validazione sperimentale. Il comportamento del materiale in funzione della frequenza di lavoro è modellato tramite complex modulus approach. L’applicazione di uno strato viscoelastico dello spessore di 1,25 mm porta a una riduzione massima del rumore pari a 9 dB.
AI-driven design and NVH analysis of automotive electric motors
Soresini, Federico
2024/2025
Abstract
This thesis focuses on Artificial Intelligence-driven design and on Noise, Vibration, and Harshness (NVH) analysis of automotive electric motors. The first objective of this research is to introduce Artificial Intelligence (AI) techniques to support and enhance the development of electric traction motors. The performance of the motor must be ensured both during the design phase and throughout its entire operational lifespan. Initially, AI methods were adopted to implement a multi-objective optimal design process for automotive electric motors. In particular, Artificial Neural Networks (ANNs) were employed to globally approximate the motor performance simulated by a Finite Element (FE) model. These surrogate models, combined with a newly developed optimization algorithm, the Adaptive Pareto Algorithm (APA), based on the iterative definition of the Pareto front, were used to optimize traction electric motors. The accuracy of the ANN was progressively improved through multiple infill operations guided by the APA, achieving a high level of predictive accuracy. The integration of Artificial Intelligence significantly increased the efficiency of the optimization process by reducing computational time, enabling the exploration of a broader design space and generating solutions that outperform those obtained through traditional design methods. Electric motor optimization was also performed within a robust design framework. AI techniques were further leveraged during the monitoring phase to detect faults in Permanent Magnet Synchronous Motors (PMSMs) under durability testing. Autoencoders (AE) were adopted as unsupervised learning techniques for this task, due to their ability to automatically extract features and patterns from operational data. After a comparative analysis of different Autoencoder architectures, a one-dimensional Autoencoder Convolutional Neural Network (1D AE-CNN) was selected as the most effective solution. Based on NVH data, the 1D AE-CNN was employed as a fault detection strategy for PMSMs operating under heavy load cycles characterized by high speed and torque variations. The implemented fault detection algorithm, tested on real motor operational data, successfully identified various PMSM faults, surpassing the limitations of traditional diagnostic methods. The AI-based fault detection system supports proactive maintenance, thereby reducing the risk of critical failures and enhancing overall system reliability. The second objective of this thesis is to study the NVH characteristics of PMSMs to develop a comprehensive understanding of their vibroacoustic behavior and guide the design of effective noise countermeasures. Electric traction motors play a major role in the overall noise emissions of electric vehicles. Moreover, they produce high-frequency tonal noise, which is particularly unpleasant for passengers. This highlights the need for a dedicated study to ensure both acoustic comfort and sound quality. To this end, an overview of current NVH challenges in the design of PMSMs for automotive applications has been carried out. Particular attention is devoted to the analytical and numerical methods used in NVH modeling, which consists of three main steps: electromagnetic force computation, motor structural modeling and acoustic emission analysis. Special focus is also placed on experimental techniques for the vibroacoustic characterization of electric motors through modal analysis. In addition to the well-established Experimental Modal Analysis (EMA) method, vibration response measurements during motor run-ups are exploited to apply Operational Deflection Shapes (ODS), Operational Modal Analysis (OMA) and Order-Based Modal Analysis (OBMA). Notably, OBMA is applied for the first time to electric motors within the context of this work. The advantages and limitations of the different techniques are compared through the study of a PMSM. The complementary use of the presented methods provides a comprehensive modal characterization of the electric motor. Finally, following a fully numerical approach, several NVH predictive models of electric motors have been employed to develop effective countermeasure techniques aimed at reducing noise emissions. The proposed solutions focus on late-design interventions that do not require modifications to the motor geometry. The first approach is based on the application of current injection strategies. This study presents an experimentally validated numerical simulation framework capable of predicting the noise reduction achievable through such strategies. The entire process is modeled, from current injection to final acoustic emission, across all simulation steps, enabling the formulation of an optimization problem for noise minimization. Using this method, a noise reduction of up to 27 dB has been achieved at the resonant frequencies of the motor. A second novel countermeasure technique to improve the NVH performance of electric machines involves applying a constrained viscoelastic rubber layer on the outer surface of a PMSM as a vibration damping treatment. The study assesses the effectiveness of this countermeasure in reducing acoustic emissions at different temperatures through a combination of numerical modeling and experimental validation. The frequency-dependent behavior of the material is accounted for using a complex modulus approach. The application of a viscoelastic layer with a thickness of 1.25 mm leads to a maximum noise reduction of 9 dB.| File | Dimensione | Formato | |
|---|---|---|---|
|
2025_07_Soresini_Federico.pdf
non accessibile
Descrizione: Tesi PhD
Dimensione
48.39 MB
Formato
Adobe PDF
|
48.39 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/241379