Currently, end-of-line bench tests on assembled transmissions represent a standard practice adopted by manufacturers to validate the overall system quality. One of the key performance requirements for modern transmissions is the vibro-acoustic behaviour. Apart from suboptimal design or assembly-related issues, the main sources of unwanted transmission noise are manufacturing defects on gears and bearings. The characteristic noise of rolling gears is defined as whine noise, particularly annoying for human hearing given its tonal characteristic. Periodic deviations on tooth flanks and pitch disrupt the conjugate engagement of gears, inducing transmission errors and increasing vibroacoustic excitation of the system. These defects, if not identified promptly, result in the rejection of the entire transmission in End-of-Line tests. Although bench tests allow for an accurate characterization of the vibro-acoustic behaviour of the transmission under real operating conditions, non-conformity detected at this stage leads to significant economic losses, resulting from reduced productivity targets, transmissions disassembly and root causes identification, scrap or rework of defective components and finally corrective actions, often with limited effectiveness due to the weak integration between end-of-line systems and upstream manufacturing processes. On the last years, the automotive sector has paid growing attention to the reduction of acoustic emissions from transmission systems. Firstly, European noise regulations (Regulation (EU) No 540/2014) impose increasingly stringent limits on vehicle noise levels, encouraging manufacturers to refine both the design and the control of mechanical noise sources. Secondly, the widespread adoption of hybrid and electric vehicles has changed the acoustic landscape: the absence of the internal combustion engine, which once masked part of the transmission noise, now makes mechanical noise more perceptible. Furthermore, the higher rotational speeds and tighter tolerances required by these systems increase sensitivity to even minor defects. Finally, the acoustic perception of a vehicle has become a distinctive attribute of the driving experience and a key indicator of perceived quality. As a result, vibro-acoustic control and characterization of the transmission have become strategic objectives to ensure comfort, performance, and product competitiveness. Automotive transmission manufacturers widely adopt hard finishing processes on gear teeth using abrasive tools such as grinding and honing to correct distortions caused by heat treatments, refine tooth flank with respect to the theoretical involute profile, and improve surface finish. Among them, the power honing process provides an efficient and cost-effective solution to meet these requirements. However, in power honing, random deviations in the quality of the pre-machined gear and dynamic instabilities may compromise the vibro-acoustic quality of the finished gears, leading to end-of-line rejects. Statistical Process Control methods are widely applied to identify quality deviations and as overall indicators of quality levels through geometric measurements of gears parameters according to technical standards. While reliable, these approaches are time-consuming, require substantial investment, and may not always represent the effective quality of the entire population, particularly in detecting defects associated with transient phenomena, such as dynamic instabilities in machine tools. In this context, the most effective strategy to prevent such issues is to detect machine fault conditions and defective components directly during machining process. Monitoring of process quantities emerges as an effective and robust solution to identify in-process both machine tools instability and deviations on the final component quality. In this work, a process monitoring model integrated with artificial intelligence has been developed based on the acquisition of vibrations through accelerometers installed on machine components, to intercept in-process components that do not comply with the target requirements. The model was trained on a dataset collected during an extensive experimental campaign conducted in an industrial context. Different feature extraction models have been adopted to compare performance; results show that the proposed model is able to predict with high accuracy the final quality of the components and to identify the associated dynamic process instabilities. It was also shown that the model can guarantee reliable accuracy even when using a small number of training samples, a key feature for effective industrial implementation.
Attualmente, i test al banco di fine linea sulle trasmissioni assemblate rappresentano una pratica standard adottata dai produttori per convalidare la qualità complessiva del sistema. Uno dei requisiti prestazionali chiave per le trasmissioni moderne è il comportamento vibroacustico. Oltre a problemi di progettazione o assemblaggio non ottimali, le principali fonti di rumore indesiderato nella trasmissione sono difetti di fabbricazione su ingranaggi e cuscinetti. Il rumore caratteristico degli ingranaggi che rotolano è definito come rumore lamentoso, particolarmente fastidioso per l'udito umano data la sua caratteristica tonale. Deviazioni periodiche sui fianchi dei denti e sul passo interrompono l'innesto coniugato degli ingranaggi, inducendo errori di trasmissione e aumentando l'eccitazione vibroacustica del sistema. Questi difetti, se non identificati tempestivamente, comportano lo scarto dell'intera trasmissione nei test di fine linea. Sebbene i test al banco consentano una caratterizzazione accurata del comportamento vibro-acustico della trasmissione in condizioni operative reali, le non conformità rilevate in questa fase comportano significative perdite economiche, derivanti da obiettivi di produttività ridotti, smontaggio delle trasmissioni e identificazione delle cause profonde, scarto o rilavorazione di componenti difettosi e, infine, azioni correttive, spesso con efficacia limitata a causa della scarsa integrazione tra i sistemi di fine linea e i processi produttivi a monte. Negli ultimi anni, il settore automobilistico ha prestato crescente attenzione alla riduzione delle emissioni acustiche dei sistemi di trasmissione. In primo luogo, le normative europee sul rumore (Regolamento (UE) n. 540/2014) impongono limiti sempre più stringenti ai livelli di rumorosità dei veicoli, incoraggiando i produttori a perfezionare sia la progettazione che il controllo delle sorgenti di rumore meccanico. In secondo luogo, l'adozione diffusa di veicoli ibridi ed elettrici ha cambiato il panorama acustico: l'assenza del motore a combustione interna, che un tempo mascherava parte del rumore della trasmissione, ora rende il rumore meccanico più percepibile. Inoltre, le maggiori velocità di rotazione e le tolleranze più strette richieste da questi sistemi aumentano la sensibilità anche a difetti minori. Infine, la percezione acustica di un veicolo è diventata un attributo distintivo dell'esperienza di guida e un indicatore chiave della qualità percepita. Di conseguenza, il controllo vibroacustico e la caratterizzazione della trasmissione sono diventati obiettivi strategici per garantire comfort, prestazioni e competitività del prodotto. I produttori di trasmissioni per autoveicoli adottano ampiamente processi di finitura dura sui denti degli ingranaggi utilizzando utensili abrasivi come la rettifica e la levigatura per correggere le distorsioni causate dai trattamenti termici, affinare il fianco del dente rispetto al profilo evolvente teorico e migliorare la finitura superficiale. Tra questi, il processo di levigatura meccanica offre una soluzione efficiente ed economica per soddisfare questi requisiti. Tuttavia, nella levigatura meccanica, deviazioni casuali nella qualità dell'ingranaggio pre-lavorato e instabilità dinamiche possono compromettere la qualità vibroacustica degli ingranaggi finiti, portando a scarti a fine linea. I metodi di Controllo Statistico di Processo sono ampiamente applicati per identificare deviazioni di qualità e come indicatori generali dei livelli di qualità attraverso misurazioni geometriche dei parametri degli ingranaggi secondo gli standard tecnici. Sebbene affidabili, questi approcci richiedono molto tempo, investimenti sostanziali e potrebbero non sempre rappresentare la qualità effettiva dell'intera popolazione, in particolare nel rilevamento di difetti associati a fenomeni transitori, come le instabilità dinamiche nelle macchine utensili. In questo contesto, la strategia più efficace per prevenire tali problemi è quella di rilevare condizioni di guasto della macchina e componenti difettosi direttamente durante il processo di lavorazione. Il monitoraggio delle grandezze di processo emerge come una soluzione efficace e robusta per identificare sia l'instabilità in-process delle macchine utensili sia le deviazioni sulla qualità del componente finale. In questo lavoro, è stato sviluppato un modello di monitoraggio di processo integrato con l'intelligenza artificiale basato sull'acquisizione di vibrazioni tramite accelerometri installati sui componenti della macchina, per intercettare i componenti in-process che non soddisfano i requisiti target. Il modello è stato addestrato su un set di dati raccolto durante un'ampia campagna sperimentale condotta in un contesto industriale. Diversi modelli di estrazione di caratteristiche sono stati adottati per confrontare le prestazioni; i risultati mostrano che il modello proposto è in grado di prevedere con elevata accuratezza la qualità finale dei componenti e di identificare le instabilità dinamiche di processo associate. È stato inoltre dimostrato che il modello può garantire un'accuratezza affidabile anche quando si utilizza un numero ridotto di campioni di addestramento, una caratteristica fondamentale per un'implementazione industriale efficace.
Dynamic vibration-based process monitoring of power-honed gears quality for automotive transmissions with an AI-integrated system / Capurso, Massimo. - ELETTRONICO. - (2025).
Dynamic vibration-based process monitoring of power-honed gears quality for automotive transmissions with an AI-integrated system
CAPURSO, MASSIMO
2025
Abstract
Currently, end-of-line bench tests on assembled transmissions represent a standard practice adopted by manufacturers to validate the overall system quality. One of the key performance requirements for modern transmissions is the vibro-acoustic behaviour. Apart from suboptimal design or assembly-related issues, the main sources of unwanted transmission noise are manufacturing defects on gears and bearings. The characteristic noise of rolling gears is defined as whine noise, particularly annoying for human hearing given its tonal characteristic. Periodic deviations on tooth flanks and pitch disrupt the conjugate engagement of gears, inducing transmission errors and increasing vibroacoustic excitation of the system. These defects, if not identified promptly, result in the rejection of the entire transmission in End-of-Line tests. Although bench tests allow for an accurate characterization of the vibro-acoustic behaviour of the transmission under real operating conditions, non-conformity detected at this stage leads to significant economic losses, resulting from reduced productivity targets, transmissions disassembly and root causes identification, scrap or rework of defective components and finally corrective actions, often with limited effectiveness due to the weak integration between end-of-line systems and upstream manufacturing processes. On the last years, the automotive sector has paid growing attention to the reduction of acoustic emissions from transmission systems. Firstly, European noise regulations (Regulation (EU) No 540/2014) impose increasingly stringent limits on vehicle noise levels, encouraging manufacturers to refine both the design and the control of mechanical noise sources. Secondly, the widespread adoption of hybrid and electric vehicles has changed the acoustic landscape: the absence of the internal combustion engine, which once masked part of the transmission noise, now makes mechanical noise more perceptible. Furthermore, the higher rotational speeds and tighter tolerances required by these systems increase sensitivity to even minor defects. Finally, the acoustic perception of a vehicle has become a distinctive attribute of the driving experience and a key indicator of perceived quality. As a result, vibro-acoustic control and characterization of the transmission have become strategic objectives to ensure comfort, performance, and product competitiveness. Automotive transmission manufacturers widely adopt hard finishing processes on gear teeth using abrasive tools such as grinding and honing to correct distortions caused by heat treatments, refine tooth flank with respect to the theoretical involute profile, and improve surface finish. Among them, the power honing process provides an efficient and cost-effective solution to meet these requirements. However, in power honing, random deviations in the quality of the pre-machined gear and dynamic instabilities may compromise the vibro-acoustic quality of the finished gears, leading to end-of-line rejects. Statistical Process Control methods are widely applied to identify quality deviations and as overall indicators of quality levels through geometric measurements of gears parameters according to technical standards. While reliable, these approaches are time-consuming, require substantial investment, and may not always represent the effective quality of the entire population, particularly in detecting defects associated with transient phenomena, such as dynamic instabilities in machine tools. In this context, the most effective strategy to prevent such issues is to detect machine fault conditions and defective components directly during machining process. Monitoring of process quantities emerges as an effective and robust solution to identify in-process both machine tools instability and deviations on the final component quality. In this work, a process monitoring model integrated with artificial intelligence has been developed based on the acquisition of vibrations through accelerometers installed on machine components, to intercept in-process components that do not comply with the target requirements. The model was trained on a dataset collected during an extensive experimental campaign conducted in an industrial context. Different feature extraction models have been adopted to compare performance; results show that the proposed model is able to predict with high accuracy the final quality of the components and to identify the associated dynamic process instabilities. It was also shown that the model can guarantee reliable accuracy even when using a small number of training samples, a key feature for effective industrial implementation.| File | Dimensione | Formato | |
|---|---|---|---|
|
38-CAPURSO Massimo.pdf
accesso aperto
Tipologia:
Tesi di dottorato
Licenza:
Creative commons
Dimensione
5.05 MB
Formato
Adobe PDF
|
5.05 MB | Adobe PDF | Visualizza/Apri |
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

