Engineering Science, School of

Receive updates for this collection

Residential Power Forecasting Based on Affinity Aggregation Spectral Clustering

Peer reviewed: 
Yes, item is peer reviewed.
Date created: 
2020-05-27
Abstract: 

Power utility companies rely on forecasting to anticipate future consumption needs, plan power production, and schedule the selling/purchasing of power. We present a novel method to forecast the power consumption of a single house based on non-intrusive load monitoring (NILM) and affinity aggregation spectral clustering, with the idea of extending it to forecasting consumption in a larger set of houses like a microgrid. First, we use a graph to model statistical relationships between appliances. Specifically, the ON/OFF time-of-day and state duration probabilities are used to compute graph edge weights and establish statistical relationships among appliances. Then, leveraging on our previous work on NILM, we disaggregate the smart meter aggregate power profile into individual appliance power profiles. With the disaggregated individual power profiles and the corresponding ON/OFF time-of-day and state duration probabilities, we next propose a method to forecast each appliance’s power profiles using affinity aggregation spectral clustering. For the proposed method, we incorporate human behaviour and environmental influence in terms of calendar and seasonal contexts in order to enhance the forecasting performance. Finally, the results of appliance-level forecasting are aggregated to perform house-level forecasting. To test our proposed forecasting method, we use four publicly available datasets and compare our method against several existing approaches such as autoregressive integrated moving average, similar profile load forecast, artificial neural network, and recent NILM-based forecasting. Experimentally, we examine how well the proposed forecasting method can generalize appliance behaviours from one house to another. Results clearly show that our method is more accurate than existing approaches.

Document type: 
Article
File(s): 

Estimating Exerted Hand Force via Force Myography to Interact with a Biaxial Stage in Real-Time by Learning Human Intentions: A Preliminary Investigation

Peer reviewed: 
Yes, item is peer reviewed.
Date created: 
2020-04-08
Abstract: 

Force myography (FMG) signals can read volumetric changes of muscle movements, while a human participant interacts with the environment. For collaborative activities, FMG signals could potentially provide a viable solution to controlling manipulators. In this paper, a novel method to interact with a two-degree-of-freedom (DoF) system consisting of two perpendicular linear stages using FMG is investigated. The method consists in estimating exerted hand forces in dynamic arm motions of a participant using FMG signals to provide velocity commands to the biaxial stage during interactions. Five different arm motion patterns with increasing complexities, i.e., “x-direction”, “y-direction”, “diagonal”, “square”, and “diamond”, were considered as human intentions to manipulate the stage within its planar workspace. FMG-based force estimation was implemented and evaluated with a support vector regressor (SVR) and a kernel ridge regressor (KRR). Real-time assessments, where 10 healthy participants were asked to interact with the biaxial stage by exerted hand forces in the five intended arm motions mentioned above, were conducted. Both the SVR and the KRR obtained higher estimation accuracies of 90–94% during interactions with simple arm motions (x-direction and y-direction), while for complex arm motions (diagonal, square, and diamond) the notable accuracies of 82–89% supported the viability of the FMG-based interactive control.

Document type: 
Article
File(s): 

Tensor Completion Methods for Collaborative Intelligence

Peer reviewed: 
Yes, item is peer reviewed.
Date created: 
2020-02-28
Abstract: 

In the race to bring Artificial Intelligence (AI) to the edge, collaborative intelligence has emerged as a promising way to lighten the computation load on edge devices that run applications based on Deep Neural Networks (DNNs). Typically, a deep model is split at a given layer into edge and cloud sub-models. The deep feature tensor produced by the edge sub-model is transmitted to the cloud, where the remaining computationally intensive workload is performed by the cloud sub-model. The communication channel between the edge and cloud is imperfect, which will result in missing data in the deep feature tensor received at the cloud side, an issue that has mostly been ignored by existing literature on the topic. In this paper we study four methods for recovering missing data in the deep feature tensor. Three of the studied methods are existing, generic tensor completion methods, and are adapted here to recover deep feature tensor data, while the fourth method is newly developed specifically for deep feature tensor completion. Simulation studies show that the new method is 3–18 times faster than the other three methods, which is an important consideration in collaborative intelligence. For VGG16’s sparse tensors, all methods produce statistically equivalent classification results across all loss levels tested. For ResNet34’s non-sparse tensors, the new method offers statistically better classification accuracy (by 0.25%–6.30%) compared to other methods for matched execution speeds, and second-best accuracy among the four methods when they are allowed to run until convergence.

Document type: 
Article
File(s): 

Human Blastocyst's Zona Pellucida Segmentation via Boosting Ensemble of Complementary Learning

Peer reviewed: 
Yes, item is peer reviewed.
Date created: 
2018-10-25
Abstract: 

Characteristics of Zona Pellucida (ZP), particularly its thickness, is a key indicator of human blastocyst (day-5embryo) quality. Therefore, ZP segmentation is an important step towards achieving automatic embryo qualityassessment. In this paper, a novel approach based on boosting ensemble of hybrid complementary learning isproposed to segment Zona Pellucida in human blastocyst images. First, an inner-ZP localization method isproposed to separate the ZP from the heavily textured area inside a blastocyst. Then, a deep Hierarchical NeuralNetwork (HiNN) is proposed to segment ZP area. The hierarchical nature of the proposed network enableslearning features with respect to their spatial location in the embryo. Finally, a Self-supervised Image-SpecificRefinement (SISR) strategy is proposed as a complementary step to boost the performance. The proposed systemis a hybrid approach in the sense that the HiNN learns the intra-correlation among images, while the SISR takesinto account the inter-correlation within the query image. Experimental results confirm that the proposed method is capable of identifying ZP area with average Precision, Recall, Accuracy and Jaccard Index of 85.2%, 92.0%, 95.6% and 78.1%, respectively. The proposed HiNN system outperforms state of the art by 4.9% in Precision, 11.2% in Recall, 3.6% in Accuracy and 10.7% in Jaccard Index.

Document type: 
Article
File(s): 

Regressing Grasping Using Force Myography: An Exploratory Study

Peer reviewed: 
Yes, item is peer reviewed.
Date created: 
2018-10-23
Abstract: 

Background: Partial hand amputation forms more than 90% of all upper limb amputations. This amputation has a notable efect on the amputee’s life. To improve the quality of life for partial hand amputees diferent prosthesis options, including externallypowered prosthesis, have been investigated. The focus of this work is to explore force myography (FMG) as a technique for regressing grasping movement accompanied by wrist position variations. This study can lay the groundwork for a future investigation of FMG as a technique for controlling externally-powered prostheses continuously. Methods: Ten able-bodied participants performed three hand movements while their wrist was fxed in one of six predefned positions. The angle between Thumb and Index fnger (θTI), and Thumb and Middle fnger (θTM) were calculated as measures of grasping movements. Two approaches were examined for estimating each angle: (i) one regression model, trained on data from all wrist positions and hand movements; (ii) a classifer that identifed the wrist position followed by a separate regression model for each wrist position. The possibility of training the system using a limited number of wrist positions and testing it on all positions was also investigated. Results: The frst approach had a correlation of determination (R2) of 0.871 for θTI and R2 θTM = 0.941. Using the second approach R2 θTI = 0.874 and R2 θTM = 0.942 were obtained. The frst approach is over two times faster than the second approach while having similar performance; thus the frst approach was selected to investigate the efect of the wrist position variations. Training with 6 or 5 wrist positions yielded results which were not statistically signifcant. A statistically signifcant decrease in performance resulted when less than fve wrist positions were used for training. Conclusions: The results indicate the potential of FMG to regress grasping movement, accompanied by wrist position variations, with a regression model for each angle. Also, it is necessary to include more than one wrist position in the training phase.

Document type: 
Article
File(s): 

Simplifying Through-Forest Propagation Modelling

Peer reviewed: 
Yes, item is peer reviewed.
Date created: 
2020-01-23
Abstract: 

Propagation analysis and modeling is critical for radio systems design, but remains a challenge for most through-vegetation situations, including forests. Transmission through such inhomogeneous mixed media is complicated by the many different propagation mechanisms and the complexity of the randomness. This means that accurate, purely physics-based analysis is unlikely to be practical (conveniently computed), and similarly, that practical, purely random modeling is unlikely to be accurate. Through-vegetation propagation models, including the standard radiative energy transfer (RET), are not very accurate in the sense that the uncertainty can be tens of dB, and this seems to be an accepted limitation for vegetation. A simpler propagation model, which maintains or improves accuracy, but keeps a reasonable association with the physics, would be insightful. This paper discusses such a model. It comprises two parallel transmission mechanisms: direct transmission through a succession of trees, which is modeled by a simple linear transmission line; and transmission across the forest top, which is modeled by simplified multiple-edge diffraction. The model is examined using recently-published experiments over a long path-length. It is demonstrated that this two-mechanism model can provide an accurate fit to the dual-slope profile of through-forest propagation over a long distance which is not possible with the RET model.

Document type: 
Article
File(s): 

The Data Gap in Sports Analytics and How to Close It

Peer reviewed: 
No, item is not peer reviewed.
Date created: 
2019-11-19
Abstract: 

As the importance and prevalence of sports analytics grows, so does the inequality in sports data. In this paper we examine two main sources of such disparity - the perceived hierarchy of sports and privatization of data. We argue that such inequality hurts the sports analytics community in the short and long terms, and suggest ways for the deep-learning, AI, and sports analytics communities to help mitigate the issue. Keywords: Sports Analytics; AI; Team Sports; Diversity

Document type: 
Article

Perceptions of Existing Wearable Robotic Devices for Upper Extremity and Suggestions for Their Development: Findings From Therapists and People With Stroke

Peer reviewed: 
Yes, item is peer reviewed.
Date created: 
2018-05-15
Abstract: 

Background: Advances in wearable robotic technologies have increased the potential of these devices for rehabilitation and as assistive devices. However, the utilization of these devices is still limited and there are questions regarding how well these devices address users’ (therapists and patients) needs.

Objective: The aims of this study were to (1) describe users’ perceptions about existing wearable robotic devices for the upper extremity; (2) identify if there is a need to develop new devices for the upper extremity and the desired features; and (3) explore obstacles that would influence the utilization of these new devices.

Methods: Focus groups were held to collect data. Data were analyzed thematically.

Results: A total of 16 participants took part in the focus group discussions. Our analysis identified three main themes: (1) “They exist, but...” described participants’ perceptions about existing devices for upper extremity; (2) “Indeed, we need more, can we have it all?” reflected participants’ desire to have new devices for the upper extremity and revealed heterogeneity among different participants; and (3) “Bumps on the road” identified challenges that the participants felt needed to be taken into consideration during the development of these devices.

Conclusions: This study resonates with previous research that has highlighted the importance of involving end users in the design process. The study suggests that having a single solution for stroke rehabilitation or assistance could be challenging or even impossible, and thus, engineers should clearly identify the targeted stroke population needs before the design of any device for the upper extremity.

Document type: 
Article
File(s): 

A Wearable Gait Phase Detection System Based on Force Myography Techniques

Peer reviewed: 
Yes, item is peer reviewed.
Date created: 
2018-04-21
Abstract: 

(1) Background: Quantitative evaluation of gait parameters can provide useful information for constructing individuals’ gait profile, diagnosing gait abnormalities, and better planning of rehabilitation schemes to restore normal gait pattern. Objective determination of gait phases in a gait cycle is a key requirement in gait analysis applications; (2) Methods: In this study, the feasibility of using a force myography-based technique for a wearable gait phase detection system is explored. In this regard, a force myography band is developed and tested with nine participants walking on a treadmill. The collected force myography data are first examined sample-by-sample and classified into four phases using Linear Discriminant Analysis. The gait phase events are then detected from these classified samples using a set of supervisory rules; (3) Results: The results show that the force myography band can correctly detect more than 99.9% of gait phases with zero insertions and only four deletions over 12,965 gait phase segments. The average temporal error of gait phase detection is 55.2 ms, which translates into 2.1% error with respect to the corresponding labelled stride duration; (4) Conclusions: This proof-of-concept study demonstrates the feasibility of force myography techniques as viable solutions in developing wearable gait phase detection systems.

Document type: 
Article
File(s):