This work aims to improve the quality control of gasoline direct injection pumps through the application of artificial intelligence on a dedicated test bench. First, the forecasting capability of the proposed neural architectures are evaluated by comparing the predictions with the experimental measurement of the flow rate delivered by the pump. The second part of the herein study is focused on evaluating the possibility of replacing a physical sensor (i.e torque meter), since, due to resonance phenomenon, its breakage has already led the partner company to extensive damage to the test bench. Among the tested structures, a Nonlinear Autoregressive Exogenous (NARX) architecture and a Long Short-Term Memory (LSTM), respectively, proved to be valid alternative to the physical sensors. At the range of interest, the neural structures showed, with respect to the experimental data, percentage errors always lower than the limit imposed by the manufactures, i.e. equals to 10%. Preliminary activities allowed to evaluate the impact of the single measured quantities and to optimize the performance of the proposed artificial architectures. All this features potentially allow to reduce and/or eliminate unnecessary physical sensors thus reducing costs and operating times.

From real to virtual sensors, an artificial intelligence approach for the industrial phase of end-of-line quality control of GDI pumps

Petrucci, L
Software
;
Ricci, F
Validation
;
Mariani, F
Methodology
;
2022

Abstract

This work aims to improve the quality control of gasoline direct injection pumps through the application of artificial intelligence on a dedicated test bench. First, the forecasting capability of the proposed neural architectures are evaluated by comparing the predictions with the experimental measurement of the flow rate delivered by the pump. The second part of the herein study is focused on evaluating the possibility of replacing a physical sensor (i.e torque meter), since, due to resonance phenomenon, its breakage has already led the partner company to extensive damage to the test bench. Among the tested structures, a Nonlinear Autoregressive Exogenous (NARX) architecture and a Long Short-Term Memory (LSTM), respectively, proved to be valid alternative to the physical sensors. At the range of interest, the neural structures showed, with respect to the experimental data, percentage errors always lower than the limit imposed by the manufactures, i.e. equals to 10%. Preliminary activities allowed to evaluate the impact of the single measured quantities and to optimize the performance of the proposed artificial architectures. All this features potentially allow to reduce and/or eliminate unnecessary physical sensors thus reducing costs and operating times.
2022
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11391/1536673
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 11
  • ???jsp.display-item.citation.isi??? 10
social impact