Borriello, Pasquale (2024) Integrating Physics-Based and Data-Driven Tools: A Methodological Framework for Design, Simulation, and Prognosis Strategies in High-Efficiency Electric Positive Displacement Pumps. [Tesi di dottorato]

Preview

Text Borriello_Pasquale_36.pdf Download (74MB) | Preview

Collection description

Positive displacement machines are integral components in various industries, spanning from industrial robotics to heavy construction equipment, aviation, and automotive sectors. Renowned for their efficiency, compactness, and reliability, these machines face new challenges in light of recent electrification trends and the pursuit of increased power density. To maintain efficiency and reliability across diverse operating conditions, leveraging virtual simulations in the design process is crucial, offering cost and time saving benefits. Virtual simulations play a pivotal role in reducing the necessity for physical prototypes, particularly during the initial stages of development. They enable a comprehensive exploration of the design space, thereby mitigating the time to market, a critical factor for manufacturers operating in a dynamic and evolving market landscape. In parallel, the advent of electric vehicles has extended operating times for such machines, incorporating challenges related to reliability even due to prolonged usage for charging intervals. Additionally, contemporary trends mandate these machines to be interconnected, and capable of transmitting critical information. This thesis addresses these evolving industrial needs by proposing rapid and efficient tools to support the design and simulation of high-efficiency, and low-noise volumetric pumps tailored for the mobility market. Furthermore, it aims to develop innovative methodologies enabling prognosis and health management strategies for commercial electric pumps. These objectives are pursued through the development and integration of both physics-based and data-driven tools within an innovative framework. In particular, a novel framework has been developed to design and simulate both the volumetric and mechanical performances of twin-spindle pumps. This modular approach, based on lumped parameters and analytical methods, has been proposed for the first time for spindle machines and addresses the market’s demand for a fast and efficient tool capable of managing the entire process from geometry generation to performance prediction. It overcomes a key limitation observed in the literature by not simplifying the geometry of the rotors when employing a lumped parameter approach. Additionally, Computational Fluid Dynamics (CFD) models employing both dynamic and static meshes have been constructed and utilized to compare performances across a wide range of operating conditions. The dynamic mesh CFD models, for the first time handled within one commercial software, have enabled the investigation of transient phenomena such as cavitation and pressure ripple. Furthermore, the accuracy of pressure ripple predictions has been validated against experimental results, and a numerical investigation has been carried out to assess the fluid-born noise associated with these machines. Concerning the topic of Prognosis and Health management, the discussion is bifurcated to address the two primary aspects: fault detection and prognosis prediction. For fault detection, two main frameworks are proposed. The first introduces a novel methodology that overcomes the primary limitation of purely data-driven approaches. This method combines an innovative knowledge-based vibroacoustic tool with prior knowledge of the kinematic chain and machine learning techniques to extract discriminatory features from time series data of pressure and acceleration signals. Notably, this approach considers defects related to both the pump head and the electric motor, including the possibility of simultaneous defects. To validate this methodology, several prototypes were specifically constructed, and an experimental test campaign was designed using Design of Experiments. The results demonstrate the critical role of specific features extracted through innovative techniques in defect detection. Overall, the proposed methodology achieves an accuracy exceeding 96%, while dimensionality reduction techniques coupled with sensor sensitivity analysis exhibit comparable efficiency while conserving 83.8% of computational time, making this procedure scalable where computational resources are limited. The second methodology for fault detection extends the previously outlined framework by integrating data from a state-of-the-art model-based simulation tool. This tool aims to diminish reliance on physical prototypes and provides the flexibility to explore a broader range of design spaces. Specifically, this methodology has been validated for detecting common gear manufacturing errors, including defects related to gear conicity, concentricity, and axial run-out. Dedicated prototypes were constructed, and an experimental campaign was conducted to validate the methodology. Pressure ripple was utilized to validate the numerical model and the implementation of concentricity and axial run-out defects has been proposed for the first time. Features extracted from pressure ripple constitute the hybrid data frame and the hybrid methodology achieved an accuracy exceeding 92% across two distinct operating conditions, demonstrating the robustness of the proposed methodology. In the realm of prognosis, a novel methodology has been proposed, incorporating several key pillars to address the predominant limitations observed in the existing literature. This approach stands out as it removes the need for a full-life test, accommodates randomness in the degradation process by providing stochastic output, offers scalability across diverse operating conditions, and facilitates design alterations by analyzing wear from erosion due to contaminants. The analysis of experimental results reveals that the evolution of pertinent variables follows a non-monotonic trajectory throughout the degradation process. Successfully modeled as a Brownian process, the degradation path is accurately depicted, with Monte Carlo simulations utilized to establish confidence intervals for each pump’s failure over time. Experimental results align closely with the predicted confidence intervals, affirming the efficacy of the proposed methodology. The scalability on different operating conditions has been validated demonstrating that the functional relationship considering speed and pressure fits the power law, as proposed in the literature. Furthermore, to predict the extent of wear attributable to erosion induced by contaminant presence within the fluid, a CFD simulation model was developed and validated against experimental data. Incorporating a dispersed phase to account for the presence of contaminants, the results highlight the tool’s potential in identifying areas most susceptible to erosion. Qualitative validation with experiments further strengthens the credibility of the approach. Such a tool could lead to design changes aimed at establishing a new design that could be less sensitive to contaminants. Ultimately, the knowledge developed through the preceding investigations has been leveraged to develop a Proof of Concept (PoC) for a Smart Pump capable of real-time monitoring of a generic circuit across six distinct operating conditions, employing absorbed current and accelerations. This PoC was prominently showcased at an international trade fair, demonstrating its potential for practical application and gaining significant attention within the industry. In conclusion, with a strong emphasis on industrial requirements, this thesis integrates physics-based and data-driven tools within a methodological framework to innovate processes related to Electric Positive Displacement Pumps. These innovative tools hold the potential for broader impacts, as they can be scaled not only to different machines but also to entire systems, amplifying their potential across various industrial domains.

Downloads

Actions (login required)

View Item