Why a Management Academy? Why should you be interested in this Engineering Management Academy from SAE? The answer to these questions lies in the statistics highlighted by surveys of hiring managers. For example, are you aware that: 28% of internal leadership promotions fail On average, it takes six years before an individual receives any formal training after being promoted to a management position Individual contributors, who are technical experts, are usually natural candidates for promotions to management positions.
Autonomous driving is currently one of the most challenging Artificial Intelligence (AI) problems as it requires combination of state-of-the-art solutions in multiple areas including computer vision, sensor fusion, control theory and software engineering. Deep learning has been pivotal to solving some of these problems, especially in computer vision. This enabled some autonomous vehicle companies started leveraging the benefits of deep learning for creating smooth, natural, human-like motion planning systems. In particular, the plethora of driving data captured from modern cars is a key enabler for training data-driven path planning systems. , Developing deep learning-powered systems relies heavily on big and high-quality data required for training of the models, in which the intrinsic statistics of the data that the model is trained on can result in different agent behavior in different scenarios.
In engineering applications, rubber isolators are subjected to continuous alternating loads, resulting in fatigue failure. Although some theoretical models are used for the fatigue life estimation of rubber materials, they do not comprehensively consider the influences of multiple factors. In the present study, a model based on the extreme learning machine (ELM) is established to estimate fatigue life of natural rubber (NR) specimens. The mechanical load (engineering strain peak), ambient temperature (23℃, 60℃ and 90℃) and shore hardness (N45 and N50) of NR specimens are used as the input variables while the measure average fatigue life as the output variable of the ELM. The regression results and predicted life distribution of the established ELM model are encouraging. For comparison, the back propagation neural network (BPNN) model and the support vector machine (SVM) model are also implemented.
Uniaxial fatigue tests of rubber dumbbell specimens under different mean and amplitude of strain are carried out. An Extreme Learning Machine (ELM) model optimized by Dragonfly Algorithm (DA) is proposed to predict the fatigue life of rubber based on measured rubber fatigue life data. Mean and amplitude of strain and measured rubber fatigue life are taken as input variables and output variables respectively in DA-ELM model. For comparison, genetic algorithm (GA) and particle swarm optimization (PSO) are used to optimize ELM parameters, and GA-ELM and PSO-ELM models are established. The comparison results show that DA-ELM model performs better in predicting the fatigue life of rubber with least dispersion. The coefficients of determination for the training set and test set are 99.47% and 99.12%, respectively. In addition, a life prediction model equivalent strain amplitude as damage parameter is introduced to further highlight the superiority of DA-ELM model.
Machine learning algorithms are effective tools to reduce the number of engine dynamometer tests during internal combustion engine development and/or optimization. This paper provides a case study of using such a statistical algorithm to characterize the heat transfer from the combustion chamber to the environment during combustion and during the entire engine cycle. The data for building the machine learning model came from a single cylinder compression ignition engine (13.3 compression ratio) that was converted to natural-gas port fuel injection spark-ignition operation. Engine dynamometer tests investigated several spark timings, equivalence ratios, and engine speeds, which were also used as model inputs. While building the model it was found that adding the intake pressure as another model input improved model efficiency.
The fuel spray process is of utmost importance to internal combustion engine design as it determines engine performance and emissions characteristics. While designers rely on CFD for understanding of the air-fuel mixing process, there are recognized shortcomings in current CFD spray predictions, particularly under super-critical or flash-boiling conditions. In contrast, time-resolved optical spray experiments have now produced datasets for the three-dimensional liquid distribution for a wide range of operating conditions and fuels. Utilizing these detailed experimental results, we have explored a machine learning approach to prediction of fuel sprays. The ML approach for spray prediction is promising because (1) it does not require phenomenological spray models, (2) it can provide time-resolved spray data without time-stepping simulation, and (3) it is computationally faster than CFD. In this study, a pixel-regression model has been developed and applied for gasoline spray prediction.
In electric power assisted steering system (EPAS), the steering assistance torque is provided by the electric motor. The motor rating is decided based on rack force requirement which depends on the vehicle weight, steering gear ratio, wheel angles & turning circle diameter etc. The load on the EPAS motor varies with respect to the steered angles of the road wheels. The motor experiences higher load towards the road wheel lock position. Most of the steering systems used on passenger cars has rack and pinion gear with constant gear ratio (C-factor). The constant gear ratio is decided to create right balance between vehicle handling behavior and steering effort. The constant gear ratio exerts higher steering load which the EPAS motor is required to support up to road wheel lock angles and hence EPAS motor size increases. This paper presents variable gear ratio (VGR) steering system in which gear ratio varies from center towards end lock stroke of rack & pinion.
Identifying edge cases for designed algorithms is critical for functional safety in autonomous driving deployment. In order to find the feasible boundary of designed algorithms, simulations are heavily used. However, simulations for autonomous driving validation are expensive due to the requirement of visual rendering, physical simulation, and AI agents. In this case, common sampling techniques, such as Monte Carlo Sampling, become computationally expensive due to their sample inefficiency. To improve sample efficiency and minimize the number of simulations, we propose a tailored active learning approach combining the Support Vector Machine (SVM) and the Gaussian Process Regressor (GPR). The SVM learns the feasible boundary iteratively with a new sampling point via active learning. Active Learning is achieved by using the information of the decision boundary of the current SVM and the uncertainty metric calculated by the GPR.
Traffic congestion, road accidents, environmental pollution and driving stress are usually associated with road transportation in dense urban environments. The widespread deployment of automated vehicles has the potential to eliminate these issues; autonomous vehicles would allow for the optimization of traffic flow, the reduction of road accidents and environmental pollution, and the elimination of driving stress. Despite recent advancements in Automated Driving Systems (ADS), the deployment of such systems in dense urban environments still faces a challenging problem: in comparison to motorway or rural driving, urban environments contain a significantly greater number of traffic participants. This makes it difficult to verify the Safety Of The Intended Functionality (SOTIF) across the entire Operational Design Domain (ODD). One approach to solve this problem is to virtually evaluate and verify the safety of the ADS using simulation tools.
This paper benchmarks three different lithium-ion (Li-ion) battery voltage modelling approaches, a physics-based using Extended Single Particle Model (ESPM), an equivalent circuit model, and a recurrent neural network. The ESPM is the selected physics-based approach because it has the advantage of having similar complexity and computational load to the other two benchmarked models. In the ESPM, the anode and cathode are simplified to single particles, and the partial differential equations are simplified to ordinary differential equations. Hence, the required state variables are reduced, and the simulation speed is improved. The second approach is a third-order equivalent circuit model (ECM), and the third approach uses a model based on a Long Short-Term Memory Recurrent Neural Network (LSTM). A Li-ion pouch cell with 47 Ah nominal capacity is used to parameterize all the models.
Accurate battery state of charge (SOC) estimation is important for safe and reliable performance of electric vehicles (EVs). Lithium ion batteries, which are commonly used for EV applications, have strong time-varying and non-linear behavior, which makes SOC estimation challenging. In this paper, a processor on the loop (PIL) platform is used to assess the execution time and memory use of different SOC estimation algorithms. Three different SOC estimation algorithms are presented and benchmarked, including an extended Kalman filter (EKF), feedforward neural network (FNN), and a recurrent neural network with long short-term memory (RNN-LSTM). The algorithms are deployed to an NXP S32K1 microprocessor and executed in real time to assess the processor loading. The impact of the number of algorithm parameters on the performance of each algorithm is also investigated.