skip to main content

This content will become publicly available on April 17, 2024

Title: A Gradient Descent Multi-Algorithm Grid Search Optimization of Deep Learning for Sensor Fusion
Sensor fusion approaches combine data from a suite of sensors into an integrated solution that represents the target environment more accurately than that produced by an individual sensor. Deep learning (DL) based approaches can address challenges with sensor fusion more accurately than classical approaches. However, the accuracy of the selected approach can change when sensors are modified, upgraded or swapped out within the system of sensors. Historically, this can require an expensive manual refactor of the sensor fusion solution.This paper develops 12 DL-based sensor fusion approaches and proposes a systematic and iterative methodology for selecting an optimal DL approach and hyperparameter settings simultaneously. The Gradient Descent Multi-Algorithm Grid Search (GD-MAGS) methodology is an iterative grid search technique enhanced by gradient descent predictions and expanded to exchange performance measure information across concurrently running DL-based approaches. Additionally, at each iteration, the worst two performing DL approaches are pruned to reduce the resource usage as computational expense increases from hyperparameter tuning. We evaluate this methodology using an open source, time-series aircraft data set trained on the aircraft’s altitude using multi-modal sensors that measure variables such as velocities, accelerations, pressures, temperatures, and aircraft orientation and position. We demonstrate the selection of an optimal DL model and an increase of 88% in model accuracy compared to the other 11 DL approaches analyzed. Verification of the model selected shows that it outperforms pruned models on data from other aircraft with the same system of sensors.  more » « less
Award ID(s):
Author(s) / Creator(s):
Date Published:
Journal Name:
2023 IEEE International Systems Conference (SysCon)
Page Range / eLocation ID:
1 to 8
Medium: X
Sponsoring Org:
National Science Foundation
More Like this
  1. In Machine learning (ML) and deep learning (DL), hyperparameter tuning is the process of selecting the combination of optimal hyperparameters that give the best performance. Thus, the behavior of some machine learning (ML) and deep learning (DL) algorithms largely depend on their hyperparameters. While there has been a rapid growth in the application of machine learning (ML) and deep learning (DL) algorithms to Additive manufacturing (AM) techniques, little to no attention has been paid to carefully selecting and optimizing the hyperparameters of these algorithms in order to investigate their influence and achieve the best possible model performance. In this work, we demonstrate the effect of a grid search hyperparameter tuning technique on a Multilayer perceptron (MLP) model using datasets obtained from a Fused Filament Fabrication (FFF) AM process. The FFF dataset was extracted from the MakerBot MethodX 3D printer using internet of things (IoT) sensors. Three (3) hyperparameters were considered – the number of neurons in the hidden layer, learning rate, and the number of epochs. In addition, two different train-to-test ratios were considered to investigate their effects on the AM process data. The dataset consisted of five (5) dominant input parameters which include layer thickness, build orientation, extrusion temperature, building temperature, and print speed and three (3) output parameters: dimension accuracy, porosity, and tensile strength. RMSE, and the computational time, CT, were both selected as the hyperparameter performance metrics. The experimental results reveal the optimal configuration of hyperparameters that contributed to the best performance of the MLP model. 
    more » « less
  2. Abstract

    This study introduces a non-invasive approach to monitor operation and productivity of a legacy pipe bending machine in real-time based on a lightweight convolutional neural network (CNN) model and internal sound as input data. Various sensors were deployed to determine the optimal sensor type and placement, and labels for training and testing the CNN model were generated through the meticulous collection of sound data in conjunction with webcam videos. The CNN model, which was optimized through hyperparameter tuning via grid search and utilized feature extraction using Log-Mel spectrogram, demonstrated notable prediction accuracies in the test. However, when applied in a real-world manufacturing scenario, the model encountered a significant number of errors in predicting productivity. To navigate through this challenge and enhance the predictive accuracy of the system, a buffer algorithm using the inferences of CNN models was proposed. This algorithm employs a queuing method for continuous sound monitoring securing robust predictions, refines the interpretation of the CNN model inferences, and enhances prediction outcomes in actual implementation where accuracy of monitoring productivity information is crucial. The proposed lightweight CNN model alongside the buffer algorithm was successfully deployed on an edge computer, enabling real-time remote monitoring.

    more » « less
  3. Grewe, Lynne L. ; Blasch, Erik P. ; Kadar, Ivan (Ed.)
    Sensor fusion combines data from a suite of sensors into an integrated solution that represents the target environment more accurately than that produced by individual sensors. New developments in Machine Learning (ML) algorithms are leading to increased accuracy, precision, and reliability in sensor fusion performance. However, these increases are accompanied by increases in system costs. Aircraft sensor systems have limited computing, storage, and bandwidth resources, which must balance monetary, computational, and throughput costs, sensor fusion performance, aircraft safety, data security, robustness, and modularity system objectives while meeting strict timing requirements. Performing trade studies of these system objectives should come before incorporating new ML models into the sensor fusion software. A scalable and automated solution is needed to quickly analyze the effects on the system’s objectives of providing additional resources to the new inference models. Given that model-based systems engineering (MBSE) is a focus of the majority of the aerospace industry for designing aircraft mission systems, it follows that leveraging these system models can provide scalability to the system analyses needed. This paper proposes adding empirically derived sensor fusion RNN performance and cost measurement data to machine-readable Model Cards. Furthermore, this paper proposes a scalable and automated sensor fusion system analysis process for ingesting SysML system model information and RNN Model Cards for system analyses. The value of this process is the integration of data analysis and system design that enables rapid enhancements of sensor system development. 
    more » « less
  4. null (Ed.)
    Deep Neural Network (DNN) trained by the gradient descent method is known to be vulnerable to maliciously perturbed adversarial input, aka. adversarial attack. As one of the countermeasures against adversarial attacks, increasing the model capacity for DNN robustness enhancement was discussed and reported as an effective approach by many recent works. In this work, we show that shrinking the model size through proper weight pruning can even be helpful to improve the DNN robustness under adversarial attack. For obtaining a simultaneously robust and compact DNN model, we propose a multi-objective training method called Robust Sparse Regularization (RSR), through the fusion of various regularization techniques, including channel-wise noise injection, lasso weight penalty, and adversarial training. We conduct extensive experiments to show the effectiveness of RSR against popular white-box (i.e., PGD and FGSM) and black-box attacks. Thanks to RSR, 85 % weight connections of ResNet-18 can be pruned while still achieving 0.68 % and 8.72 % improvement in clean- and perturbed-data accuracy respectively on CIFAR-10 dataset, in comparison to its PGD adversarial training baseline. 
    more » « less
  5. The effectiveness of obstacle avoidance response safety systems such as ADAS, has demonstrated the necessity to optimally integrate and enhance these systems in vehicles in the interest of increasing the road safety of vehicle occupants and pedestrians. Vehicle-pedestrian clearance can be achieved with a model safety envelope based on distance sensors designed to keep a threshold between the ego-vehicle and pedestrians or objects in the traffic environment. More accurate, reliable and robust distance measurements are possible by the implementation of multi-sensor fusion. This work presents the structure of a machine learning based sensor fusion algorithm that can accurately detect a vehicle safety envelope with the use of a HC-SR04 ultrasonic sensor, SF11/C microLiDAR sensor, and a 2D RPLiDAR A3M1 sensor. Sensors for the vehicle safety envelope and ADAS were calibrated for optimal performance and integration with versatile vehicle-sensor platforms. Results for this work include a robust distance sensor fusion algorithm that can correctly sense obstacles from 0.05m to 0.5m on average by 94.33% when trained as individual networks per distance. When the algorithm is trained as a common network of all distances, it can correctly sense obstacles at the same distances on average by 96.95%. Results were measured based on the precision and accuracy of the sensors’ outputs by the time of activation of the safety response once a potential collision was detected. From the results of this work the platform has the potential to identify collision scenarios, warning the driver, and taking corrective action based on the coordinate at which the risk has been identified.

    more » « less