Load assumption process for durability design using new data sources and data analytics

https://doi.org/10.1016/j.ijfatigue.2020.106116Get rights and content

Abstract

Durability engineering for vehicles is about relating the real operational loading and fatigue environment to the actual capability and strength capacity of the product and its parts. The statistical modelling of the usage loading distribution has always been a big challenge; the latter is necessary to derive statistically validated design targets. In the past, dedicated durability measurement campaigns were the only available data source to obtain durability loads. Nowadays, this has changed due to the increasing availability of new complementary data sources. We show how such data like field monitoring data and geographic data, along with the usage of new data analytics and simulation methods can be used to derive customer and usage related durability loads and targets.

Introduction

Durability engineering for vehicles is about relating the real operational loading and fatigue environment, experienced by the vehicle over its lifetime, to the strength capacity of its structure and components.

The assessment of the actual strength and fatigue properties is done on system, subsystem and component level by tests on test tracks, on test rigs and by numerical multi-body system (MBS) and finite element (FE) based fatigue life prediction [1]. In all these cases, usually the loading schedule is set as a high quantile of the so-called customer usage loading distribution, i.e., of the distribution of the loading experienced by the vehicles over their (nominal) lifetime ([1], [2], [3]). Various versions of such durability engineering processes have been built and refined throughout the industry since the 1980s.

The biggest challenge in all those processes still is the statistical modelling of the usage loading distribution and the collection and qualification of representative data to align those statistical models with reality.

Data sources for durability loads

In this paper, we discuss three different types of data sources (termed D1, D2 and D3) as described below.

Traditionally, durability load data are collected in so-called durability measurement campaigns (D1). A vehicle, equipped with suitable, typically expensive, measurement technology is sent to a specific region where a planned campaign of typically several thousand kilometers is driven. Thus, high quality data (forces, strains, accelerations, displacements) are measured at high resolution in the relevant frequency range. A big advantage of such data is that they are directly suited as input for simulation based load cascading to derive component loads. Also deriving the correlation to test tracks (if measured with the same vehicle) is possible. A disadvantage, however, is the lack of invariance, i.e. the transfer of any findings to different vehicles or variants requires quite some extra nontrivial methods and effort. Another important shortcoming is that the measurement campaign data alone are not representative for customer usage.

Up until a couple of years ago, such measurement campaign data (D1) were almost the only available data source for durability loads. This has changed in the last years because of the increasing availability of new complementary data sources. In this paper, we discuss two kinds of such new data sources. We call them field monitoring data (D2) and geographic data (D3).

Field monitoring data (D2): Accelerations, displacements, engine-powertrain state data and other sensor data are nowadays available on the vehicles data-bus. Those data are primarily acquired to feed the advanced-driving-assistance systems (ADAS) and the engine powertrain control systems. Meanwhile, it is quite common to use such data also for condition monitoring purposes. To provide better durability load data, field monitoring data can be used in two different ways: (a) Directly, by collecting the data and transferring them either by telemetry (vehicle telematics) or in accumulated and condensed form, e.g., at inspection. Here, one faces the problem that typically the data accuracy and resolution is not sufficient for MBS and FE based durability analysis. (b) Indirectly, by using the data primarily to identify the vehicle states and maneuvers. This can be done with good quality and, if it is done for many vehicles or even for complete vehicle fleets, this process yields suitable big data for the statistics of real usage.

Geographic data (D3): Map data, road network data, including rules and signs, topography, road conditions, traffic data, land use data and weather data are nowadays available in digital formats. These data can be obtained from different sources, such as from map data providers, satellite survey programs, weather services, road authorities and more (see [4], [5], [6], [7], [8]).

Especially for vehicle engineering applications, like those discussed in this paper, the Fraunhofer so-called virtual-measurement-campaign (VMC) software provides an aggregated and qualified database of such geographic data, which makes this approach much more practical. For a detailed description we refer to [4], [5], [6], [7]. Of course, such data are not vehicle specific, they are not even vehicle related and thus they cannot be used directly to derive load data. However, in combination with appropriate vehicle simulation models they can be used to assess the influence of road types, hilliness, curviness, and even of traffic situations on the vehicle’s performance and durability. And, since the data are vehicle independent and available on big scale, different vehicle types and variants can be analyzed and optimized for different mission types in different countries efficiently.

Data analytics and simulation methods

To make good use of all these data we need appropriate mathematical models and data analytics methods to relate them to the durability performance of vehicles. At the end of the day we want the data to help us with statements like: This vehicle is safe and has a (defined) very low probability of failure for a certain usage characteristic in a certain country and yet it is not too much overdesigned. This requires predictions - and there are no predictions without mathematical models.

In this paper, we will make use of four kinds of mathematical modelling and simulation methods (denoted herein as M1 to M4):

Detailed vehicle system models (M1): Detailed vehicle system models, especially multi-body simulation models (MBS) are well established in vehicle CAE. Their main role in durability engineering is what we call load cascading, i.e., transferring loads specified on system level to subsystem loads and component loads, which, in turn, are the input for numerical (FE-based) fatigue life prediction or for component tests. System loads (input) may be given as measured wheel forces (from (D1)) or, more invariant, as digital road profiles together with a target speed profile. In the second case, the model needs to include tires and a driver. Such models are very powerful and widely used. Limiting drawbacks are simulation effort (real-time factors 50–100 are typical, i.e., simulation time may be 50 to 100 times longer than the real time of the maneuver simulated) and the difficulty to get a consistent parameterization and qualification of the model due to the complexity including flexible bodies, rubber bushings, hydro-bushings, tires and controllers.

Simplified longitudinal dynamics models (M2) are less detailed and therefore not suited for load path simulation. However, they are much faster than full MBS models (typical real-time factors: 0.1–1, or even lower). They contain the vehicles mass, power and wheelbase as well as some powertrain information and a driver model. One can also couple them with stochastic traffic models and thus simulate a vehicle and traffic dependent speed profile based on geographic data (D3). We will explain more about this method in Section 4. For more background and more application cases we refer to [5], [7], [8], [9]. Additionally, such models can be used efficiently in an inverse manner, e.g., for identification purposes. In Section 5, we demonstrate the identification of road roughness and road profiles inverting a quarter-car model and using vehicle measurements (D1), we refer to [10] as well.

Data-based prediction/machine learning (M3): Data-based models are derived from given input–output data. To be more precise, a specifically chosen model (structure), parametric or non-parametric, is trained and optimized, respectively, such that it approximates best the given input–output data. In the context of machine learning, this task is called a supervised learning problem. A prominent example class for a model structure are (artificial) neural nets, but many other classes and models are possible as well. Once a model has been trained (i.e., learned from the data), it can be used for prediction and extrapolation; that is, it is applied to new input data that have not been in the training data.

Monte Carlo type usage simulation (M4): This is about modelling and simulating the statistics of vehicle usage. The resulting statistics allows understanding and quantifying the usage variability of vehicles. The idea is to simulate thousands of vehicle life spans of, say, 300.000 km or 15.000 h of operation each. The input data for such simulations can be either geographic data (D3) like road network, topography, road conditions, traffic data, and points of interest. Alternatively, it can be properly segmented rich data from measurement campaigns (D1). The results can be statistically well qualified durability load targets, given as high quantiles of the simulated usage distribution. We will explain more about this method and its application in Section 2.

Outline of the following sections

We will present and discuss different engineering applications and processes for modelling usage variability and durability loading. We want to emphasize how the new data sources and simulation methods described above enable those processes. In Section 2, we describe a general process for the derivation of statistically validated design loads, which combines the three different data type (D1), (D2) and (D3) using the modeling and simulation models (M1) and (M4). In Section 3, we demonstrate how vehicle data can be used to reveal certain usage types. In particular, this features the indirect use of field-monitoring data (D2) combined with machine learning methods (M3). In Section 4, we discuss the use of geographic data (D3) in combination with simplified simulation models (M2), in order to predict usage- and customer-specific durability loads and energy demands. As mentioned, in Section 5, we present a methodology for the identification of road roughness and road profiles with the help of simplified models (M2), vehicle measurements (D1) and the mathematical inversion theory. Moreover, we show how to use that identified information for prediction by relating the information to geographic data (D3) and data-based extrapolation using sampling techniques (M4).

This paper is an extended version of the paper submitted for the proceedings of the Fourth International Conference on Material and Component Performance under Variable Amplitude Loading (VAL4), which was scheduled for May/April 2020 and unfortunately cancelled because of the Covid 19-virus, [11].

Section snippets

Data sources and data processing for the derivation of design loads

In this section, we consider the process of collecting and evaluating high quality load data in order to derive design loads for safety critical and functional components. Fig. 1 gives a rough overview of the process.

The traditional approach requires a database of high quality data at many spots of the vehicle (wheel forces, suspension forces, strains and accelerations, spring displacements etc.), from which we shall derive the design loads.

A very simple option for e.g. a truck would be to

Usage type detection and usage profiles based on vehicle data

In this chapter, we demonstrate how vehicle data can be used to detect certain usage types. In particular, this features the indirect use of field-monitoring data (D2) combined with machine learning methods (M3). Modern vehicles usually record much data and information routinely, and commercial vehicles are often equipped with telematics systems that send data to a cloud at regular intervals. At the same time, in the area of commercial vehicles, especially agricultural and construction

Usage-dependent prediction of loads and energy demands

In this section, we discuss the use of geographic data (D3) in combination with simplified simulation models (M2), in order to predict usage- and customer-specific energy demands. In particular, during the last years, the Fraunhofer ITWM has developed modelling and simulation methods that allow to analyze efficiently the strongly varying usage variability and the resulting variations in the loads and in the performance requirements and to consider them early in the vehicle development process.

Geo-referenced estimation and prediction of road roughness

In this section, we present and discuss an approach that identifies road profiles and road roughness data based on vehicle measurements and that allows road data prediction by relating the identified information to map and environmental data.

In the vehicle development process, it is highly important to know as much as possible about the customer-specific market situation. The vehicle is designed to meet region- and customer-specific durability requirements. Thus, it is a primary goal to collect

Summary and outlook

We have shown how new, complementary data sources, such as field monitoring data (D2) and geographic data (D3) in combination with new data analytics and simulation methods, such as simplified longitudinal dynamics models (M2), data based models (M3) and Monte Carlo type models (M4) enable new engineering processes, in particular for the derivation of durability loading targets. Usage models allow the simulation of thousands of vehicle lives according to a certain usage type. Field monitoring

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

References (35)

  • P. Johannesson et al.

    Guide to load analysis for durability in vehicle engineering

    (2013)
  • Dressler K, Speckert M, Müller R, Weber C. Customer loads correlation in truck engineering. In: Proceedings of FISITA...
  • M. Speckert et al.

    Customer usage profiles, strength requirements and test schedules in truck engineering

  • Speckert M, Dressler K, Lübke M. T. Halfmann, Automatisierte und um GEO-Daten angereicherte Auswertung von Messdaten...
  • M. Speckert et al.

    The Virtual Measurement Campaign (VMC) concept - A methodology for geo-referenced description and evaluation of environmental conditions for vehicle loads and energy efficiency

  • Dressler K, Speckert M. Simulation der Nutzungsvariabilität für Betriebsfestigkeit und Energieeffizienz unter...
  • M. Speckert et al.

    Die virtuelle Messkampagne (VMC) - ein geo-referenziertes System für die Fahrzeugauslegung hinsichtlich Beanspruchung

  • Burger M, Speckert M, Dressler K. Nutzungsspezifische Vorhersage von Beanspruchung und Energieverbrauch. In:...
  • K. Sandmann et al.

    Ableitung von feldrelevanten Lastkollektiven mittels stochastischer Verkehrssimulation

  • M. Burger et al.

    Model-Based Identification Of Road Profiles and Road Roughness Indicators Using Vehicle Measurements

  • Burger M, Dressler K, Speckert M. New Data Sources and Data Analytics Methods for Durability Engineering. In:...
  • M. Wagner et al.

    Determining Customer-Correlated Loads by Using the Monte Carlo Method

  • R. Müller et al.

    Ableitung von einsatzspezifischen Auslegungslasten der Betriebsfestigkeit - Ermittlung globaler Marktanforderungen am Beispiel des südamerikanischen Marktes

  • M. Schindler et al.

    Beanspruchungskollektive für Nutzfahrzeuge in Gesamteuropa - Ein innovativer Extrapolationsansatz

    Mater Test

    (2011)
  • Streit A, Dressler K, Speckert M, Lichter J, Zenner T, Bach P. Anwendung statistischer Methoden zur Erstellung von...
  • C. Eckstein et al.

    Determination of test scenarios for durability verification of tractors under consideration of their usage variability?

  • D. Montgomery et al.

    Applied statistics and probability for engineers

    (2003)
  • Cited by (9)

    View all citing articles on Scopus
    View full text