Leveraging Machine Learning for Early Design Decisions on an Accessory Belt Drive Simulation

There are various challenges faced by an automotive engineer while designing a robust and optimized accessory drive system. Most original equipment manufacturers (OEMs) rely on different suppliers for their engine belt(s) and accessories (e.g. water pumps, alternators, A/C compressors, etc.). This leads to challenges in obtaining a comprehensive set of input data to incorporate in a system-level simulation. In instances where the analysis is being performed by the belt supplier, there are often issues obtaining permission to share accessory input data for use in the simulation. Instead, it may be preferable for the OEMs to create their own common simulation platform and system level models using data collected from their various suppliers

In case of an issue in the accessory drive system, it is challenging to get a common model to analyze the entire system in one simulation platform. Most of the time, sharing input data of one supplier with other leads to proprietary issues. It is necessary for OEMs to create their own common simulation platform and system-level models using inputs collected from their various suppliers. 

A System-level Approach to Analyze Accessory Drives 

On many occasions, due to simulation tool limitations, crank loading on the accessory drive is modeled as an imposed speed including torsional oscillation of the crank pulley. But this approach does not consider the dynamic coupling between cranktrain and accessory drive system dynamics. GT-SUITE’s multi-physics capabilities can model the system-level interactions between the torsional crankshaft model, excited by cylinder pressure traces, and the detailed belt model, including accessories and tensioners. This integrated model helps design a robust product and allows for system-level optimization. For example, in a single model, changes in a torsional vibration damper (TVD) on a crankshaft can be analyzed from a belt dynamics isolation standpoint.  

Using GT-SUITE, it is also possible to model the accessory drive system for mild hybrid electric vehicle (HEV) applications. Mild hybrids are often fitted with a dual arm tensioner for boosting/re-generation or start/stop applications. These tensioners are easy to construct and incorporate into the system model to accommodate the change in tight and slack spans as the direction of load transfer changes during operation.  

Note: Using GT-SUITE, detailed cranktrains, timing drives, and valvetrain subsystems can be integrated to perform a detailed system-level analysis of the entire powertrain (see Figure 1 below). 

Figure 1: System-level Modeling of an Accessory Drive in GT-SUITE

Longitudinal and Transverse Vibration in Belt Drive 

In belt drive systems there are fundamentally two major vibration issues which can cause failure: 

Torsional Vibration: A belt and pulley system can be conceptualized as a rotational system of torsional springs and rotational inertias. The span between two pulleys acts as a spring and the pulleys as rotational inertias in a simplified representation. This system can get excited at its torsional natural frequency due to the crank pulley excitation. When there is an alternator decoupler in the system, which has soft springs (used to decouple the heavy inertia of the generator), it can create a first natural frequency in the range of 10-20 Hz. 

Transverse/Span Vibration: The belt segment between two pulleys has its own stiffness and can be conceptualized as string with standing wave dynamics. This span stiffness depends on span length and initial belt installation tension. The shorter the span length, the higher the stiffness and higher the span natural frequency. In many cases, due to the packaging, it is difficult to reduce the span length. In this case, installation tension becomes a critical parameter to reduce transverse vibration.

See Figure 2 to see the difference between torsional and transverse vibrations.

Figure 2: Torsional and Transverse Vibration in an Accessory Drive

Main Simulation Results to be Analyzed 

For analyzing the accessory drive system, below are the main results that need to be analyzed with defined design targets: 

• Maximum and minimum belt tension 
• Maximum and minimum hub loads 
• Tensioner arm angular motion 
• Slips at each pulley 
• Transverse belt span deflections 

Among these, maximum belt tension and tensioner arm motion are the most critical output as failing these criteria can lead to the mechanical failure of the belt. There is often a hard stop in the tensioner. Once the angular arm motion reaches this limit, then there is an abrupt increase in belt tension which may lead to breakage of the belt. Hub load outputs are also important for bearing design and durability considerations. Minimum belt tension, hub load, slip, and span deflection are important for efficient design. For example, if the minimum belt tension gets too low, the belt may lose sufficient contact load with the pulley, reducing the system’s efficiency with unwanted slip.

In Figure 3 below, here are some examples of the results in GT-SUITE (e.g., belt tension and global slip %). Note that this example was based on a genset application in which the front end auxiliary drive (FEAD) is running at single engine speed. 

Figure 3: Belt Tension and Global Slip%

Use of Metamodels to Provide Early Design Direction to the Product Team  

Once a baseline model is defined, including the ability to qualify design guidelines, constraints, and identify failing criteria, it is ready to leverage GT-SUITE’s built-in Machine Learning Assistant (MLA). The MLA can be used for following scenarios: 

1. Cost Optimization (e.g. reducing the number of ribs on the belt, removing the idler, and so forth) 
2. Robust Design (e.g. choosing the appropriate belt type either as an aramid or polyester cord, setting up optimum installation belt tension, and more for optimum performance and durability)

In the plots below (Figure 4), parameters like belt pre-tension, water pump/alternator torque, belt axial stiffness, idler diameter, and tensioner position are studied for a given model to evaluate their relative impact on critical output. In general, these parameters have some flexibility to change while designing the system. Based on the sensitivity analysis, a design direction can be given to the production team.  

Figure 4: Main Effects Plots for Specified Attributes or Inputs Ranking

Also, there is another group of plots which show variational analysis results (Figure 5) within the reasonable ranges specified for the different input parameters. This will help the production team to choose the right design. For example, in this case based on the sensitivity analysis, a user can first try to optimize the system–based on the belt dynamic tension and tensioner angular motion by changing the belt axial stiffness or water pump torque. The effect on the dynamic belt tension and tensioner motion within the range can be seen by the slider bar results as shown in Figure 5. 

Figure 5: Variational Analysis for Specified Attributes

Video Demonstration of Kriging and Multilayer Perceptron (MLP) Metamodeling Methods   

Learn More About our Accessory Drive Belt Dynamic Simulation  

If you would like to learn more or are interested in trying accessory drive belt dynamic simulation, contact us!  

Stay tuned for the next blog which will be focused on the details of the use of GT-SUITE’s Machine Learning Assistant for accessory drives!