Search Results

Self-piercing rivets (SPR) are efficient and economical joining methods used in the manufacturing of lightweight automotive bodies. The finite element method (FEM) is a potentially effective way to assess the joining process of SPRs. However, uncertain parameters could lead to significant mismatches between the FEM predictions and physical tests. Thus, a sensitivity study on critical model parameters is important to guide the high-fidelity modeling of the SPR insertion process. In this paper, an axisymmetric FEM model is constructed to simulate the insertion process of the SPR using LS-DYNA/explicit. Then, several surrogate models are evaluated and trained using machine learning methods to represent the relations between selected inputs (e.g., material properties, interfacial frictions, and clamping force) and outputs (cross-section dimensions).

The automotive industry is subject to stringent regulations in emissions and growing customer demands for better fuel consumption and vehicle performance. Engine calibration, a process that optimizes engine performance by tuning engine controls (actuators), becomes challenging nowadays due to significant increase of complexity of modern engines. The traditional sweep-based engine calibration method is no longer sustainable. To tackle the challenge, this work considers two powerful global optimization methods: genetic algorithm (GA) and Bayesian optimization for steady-state engine calibration for single speed-load point. GA is a branch of meta-heuristic methods that has shown a great potential on solving difficult problems in automotive engineering. Bayesian optimization is an efficient global optimization method that solves problems with computationally expensive testing such as hyperparameter tuning in deep neural network (DNN), engine testing, etc.

This study compares the model efficacy of Neural Network and Vector Auto Regression. Further it also analyses the impact of predictors controlling for total industry volume. Understanding both the methodologies has their distinctive advantages and disadvantages. Our empirical findings indicate that based on the characteristics of data such as non-stationary, non-linearity and non-normality paves the way for use of machine learning algorithm relative to econometrics technique. Our results suggest that data type and its characteristics are more important in determining the methodology than the methodology itself. In industry, econometrics methodologies are widely used due to their usage simplicity and its ability to explain the relationships in simple terms.

In this work, we outline a process for traffic light detection in the context of autonomous vehicles and driver assistance technology features. For our approach, we leverage the automatic annotations from virtually generated data of road scenes. Using the automatically generated bounding boxes around the illuminated traffic lights themselves, we trained an 8-layer deep neural network, without pre-training, for classification of traffic light signals (green, amber, red). After training on virtual data, we tested the network on real world data collected from a forward facing camera on a vehicle. Our new region proposal technique uses color space conversion and contour extraction to identify candidate regions to feed to the deep neural network classifier. Depending on time of day, we convert our RGB images in order to more accurately extract the appropriate regions of interest and filter them based on color, shape and size. These candidate regions are fed to a deep neural network.

This work describes a single camera based object distance estimation system. As technology on vehicles is constantly advancing on the road to autonomy, it is critical to know the locations of objects in 3D space for safe behavior of the vehicle. Though significant progress has been made on object detection in 2D sensor space from a single camera, this work additionally estimates the distance to said object without requiring stereo vision or absolute knowledge of vehicle motion. Specifically, our proposed system is comprised of three modules: vision based ego-motion estimation, object-detection, and distance estimation. In particular, we compensate for the vehicle ego-motion by using pin-hole camera model to increase the accuracy of the object distance estimation.

While operating a vehicle in either autonomous or occupant piloted mode, an array of sensors can be used to guide the vehicle including stereo cameras. The state-of-the-art distance map estimation algorithms, e.g. stereo matching, usually detect corresponding features in stereo images, and estimate disparities to compute the distance map in a scene. However, depending on the image size, content and quality, the feature extraction process can become inaccurate, unstable and slow. In contrast, we employ deep convolutional neural networks, and propose two architectures to estimate distance maps from stereo images. The first architecture is a simple and generic network that identifies which features to extract, and how to combine them in a multi-resolution framework. The second architecture is a more specialized one that extracts local similarity information from two images, which are used for stereo feature matching, and fuses them at multiple resolutions to generate the distance map.

Use of sensors to monitor dynamic performance of machine tools at Ford’s powertrain machining plants has proven to be effective. The traditional approach to convert sensor data to actionable intelligence consists of identifying single features from cycle based signatures and setting thresholds above acceptable performance limits based on trials. The thresholds are used to discriminate between acceptable and unacceptable performance during each cycle and raise alarms if necessary. This approach requires a significant amount of resource & time intensive set up work up-front and considerable trial and error adjustments. The current state does not leverage patterns that might be discernible using multiple features simultaneously. This paper describes enhanced methods for processing the data using supervised and unsupervised machine learning methods. The objective of using these methods is to improve the prediction accuracy and reduce up-front set up.

Plug-in hybrid (PHEV) technology offers the ability to achieve zero tailpipe emissions coupled with convenient refueling. Fleet adoption of PHEVs, often motivated by organizational and regulatory sustainability targets, may not always align with optimal use cases. In a car rental application, barriers to improving fuel economy over a conventional hybrid include: diminished benefits of additional battery capacity on long-distance trips, sparse electric charging infrastructure at the fleet location, lack of renter understanding of electric charging options, and a principle-agent problem where the driver accrues fewer benefits than costs for actions that improve fuel economy, like charging and eco-driving. This study uses high-resolution driving data collected from twelve Ford Fusion Energi sedans owned by University of California, Davis (UC Davis), where the vehicles are rented out for university-related activities.

This paper presents an approach of using neural network and fuzzy logic methods for the diagnosis of fault root causes in a manufacturing environment. As the first step in this approach, data from all the valid test points were collected and studied based on their statistical characteristics. An information-gain-based procedure was then followed to quantitatively evaluate the relevance of each test point to the diagnosis process. Accordingly, an objective rank of all relevant test points was generated for a particular reject. The root cause of rejects was then identified by a procedure based on an information-gain-weighted radial basis function neural network and a fuzzy multiple voting classification algorithm. This method has been tested with the top five rejects of the transmission main control component at Ford and promising results have been obtained.

During the design conception of an automobile, typically low-fidelity physics-based simulations are coupled with engineering judgement to define key architectural components and subsystems which limits the capability to identify NVH issues arising from systems interaction. This translates to non-optimal designs because of unexplored design opportunities and therefore, lost business efficiencies. The sparse design information available during the design conception phase limits the development of representative higher fidelity physics-based simulations. To address that restriction on design optimization opportunities, this paper introduces an alternate approach to develop reduced order predictive models using regression techniques by harnessing historical measurement and simulation data. The concept is illustrated using two driveline NVH phenomenon: axle whine and take-off shudder.

To enable smooth and low-risk autonomous driving in the presence of other road users, such as cyclists and pedestrians, appropriate predictive safe speed control strategies relying on accurate and robust prediction models should be employed. However, difficulties related to driving scene understanding and a wide variety of features influencing decisions of other road users significantly complexifies prediction tasks and related controls. This paper proposes a hierarchical neural network (NN)-based prediction model of pedestrian crossing behavior, which is aimed to be applied within an autonomous vehicle (AV) safe speed control strategy. Additionally, different single-level prediction models are presented and analyzed as well, to serve as baseline approaches.

Drivers often engage in secondary in-vehicle activity that is not related to vehicle control. This may be functional and/or to relieve monotony. Regardless, drivers believe they can safely do so when their perceived workload is low. In this paper, we describe a data acquisition system and machine learning based algorithms to determine perceived workload. Data collected were from on-road driving in light and heavy traffic, and individual physiological measures were recorded while the driver also performed in-vehicle tasks. Initial results show how the workload function can be personalized to an individual, and what implications this may have for vehicle design.

The proposed technique is a tailored deep neural network (DNN) training approach which uses an iterative process to support the learning of DNNs by targeting their specific misclassification and missed detections. The process begins with a DNN that is trained on freely available annotated image data, which we will refer to as the Base model, where a subset of the categories for the classifier are related to the automotive theater. A small set of video capture files taken from drives with test vehicles are selected, (based on the diversity of scenes, frequency of vehicles, incidental lighting, etc.), and the Base model is used to detect/classify images within the video files. A software application developed specifically for this work then allows for the capture of frames from the video set where the DNN has made misclassifications. The corresponding annotation files for these images are subsequently corrected to eliminate mislabels.

Plug-in hybrid electric vehicles (PHEV) have an EV mode driving range which can cover a portion of customer daily driving. This EV mode range affects the refuel frequency substantially compared with conventional vehicle. For a conventional vehicle, daily driving pattern, tank size and fuel economy are the factors affecting the refuel frequency. While for a PHEV, EV range is another factor would affect the results substantially. Traditional method of label range can’t represent real world driving range between fill-ups for PHEV well. How to accurately predict the PHEV refuel distance taking into account real world customer driving patterns and PHEV parameters become critical for PHEV system design and optimization. This paper presents real world big customer data based PHEV refuel distance estimation modeling. The target is to estimate PHEV refuel distance given several specific parameters such as EV range, hybrid mode fuel economy, tank size etc.

Self-piercing rivet (SPR) joints are a key joining technology for lightweight materials, and they have been widely used in automobile manufacturing. Manual visual crack inspection of SPR joints could be time-consuming and relies on high-level training for engineers to distinguish features subjectively. This paper presents a novel machine learning-based crack detection method for SPR joint button images. Firstly, sub-images are cropped from the button images and preprocessed into three categories (i.e., cracks, edges and smooth regions) as training samples. Then, the Artificial Neural Network (ANN) is chosen as the classification algorithm for sub-images. In the training of ANN, three pattern descriptors are proposed as feature extractors of sub-images, and compared with validation samples. Lastly, a search algorithm is developed to extend the application of the learned model from sub-images into the original button images.

All drivers come with a driving signature during a driving. By aggregating adequate driving data of a driver via multiple driving sessions, which is already embedded with driving behaviors of a driver, driver identification task could be treated as a supervised machine learning classification problem. In this paper, we use a random forest classifier to implement the classification task. Therefore, we collected many time series signals from 60 driving sessions (4 sessions per driver and 15 drivers totally) via the Controller Area Network. To reduce the redundancy of information, we proposed a method for signal pre-selection. Besides, we proposed a strategy for parameters tuning, which includes signal refinement, interval feature extraction and selection, and the segmentation of a signal. We also explored the performance of different types of arrangement of features and samples.

Electric vehicles are highly sensitive to variations in environmental factors (like temperature, drive style, grade, etc.). The distribution of real-world range of electric vehicles due to these environmental factors is an important consideration in target setting. This distribution can be obtained by running several simulations of an electric vehicle for a number of high-frequency velocity, grade, and temperature real-world trip profiles. However, in order to speed up simulation time, a unique set of drive profiles that represent the entire real-world data set needs to be developed. In this study, we consider 40,000 unique velocity and grade profiles from various real-world applications in EU. We generate metadata that describes these profiles using trip descriptor variables. Due to the large number of descriptor variables when considering second order effects, we normalize each descriptor and use principal component analysis to reduce the dimensions of our dataset to six components.

As the world searches for ways to reduce humanity’s impact on the environment, the automotive industry looks to extend the viable use of the gasoline engine by improving efficiency. One way to improve engine efficiency is through more effective control. Torque-based control is critical in modern cars and trucks for traction control, stability control, advanced driver assistance systems, and autonomous vehicle systems. Closed loop torque-based engine control systems require feedback signal(s); indicated mean effective pressure (IMEP) is a useful signal but is costly to measure directly with in-cylinder pressure sensors. Previous work has been done in torque and IMEP estimation using crankshaft acceleration and ion sensors, but these systems lack accuracy in some operating ranges and the ability to estimate cycle-cycle variation.

Rapid development of Laser Powder Bed Fusion (L-PBF) technology enables almost unconstrained design freedom for metallic parts and components in automotive industry. However, the mechanical properties of L-PBF alloys, AlSi10Mg for example, have shown significant differences when compared with their counterparts via conventional manufacturing process, due to the unique microstructure induced by extremely high heating and cooling rate. Therefore, microstructure informed material modeling approach is critical to fully unveil the process-structure-property correlation for such materials and enable the consideration of the effect of manufacturing during part design. Multi-scale material modeling approach, in which crystal plasticity finite element (CPFE) models were employed at the microscale, has been previously developed for L-PBF AlSi10Mg.

Connected vehicle data unlock compelling solutions for vehicle owners and fleet managers. In selecting machine learning algorithms for use in predicting a connected vehicle signal value, time series dependency is critical to understand. With little to no time series dependency, conventional machine learning models may be used with a feature set that has few or no lag variables. If there is a lot of time series dependency including long-term dependencies, deep learning architectures like variants of recurrent neural networks (RNN) may be a better approach. Further, at any time step, RNN features may be specified to use some number of past time steps to predict the latest value. This paper seeks to identify time series dependency of connected vehicle signals, and selection of the number of time steps to look back in the features set to minimize error.