Flexible manufacturing is one of the core goals of Industry 4.0 and brings new challenges to current industrial control systems. Our detailed field study on auto glass industry revealed that existing production lines are laborious to reconfigure, difficult to upscale, and costly to upgrade during production switching. Such inflexibility arises from the tight coupling of devices, controllers, and control tasks. In this work, we propose a new architecture for industrial control systems named Control-as-a-Service (CaaS). CaaS transfers and distributes control tasks from dedicated controllers into network switches. By combining control and transmission functions in switches, CaaS virtualizes the whole industrial network to one Programmable Logic Controller (PLC). We propose a set of techniques that realize end-to-end determinism for in-network industrial control and a joint task and traffic scheduling algorithm. We evaluate the performance of CaaS on testbeds based on real-world networked control systems. The results show that the idea of CaaS is feasible and effective, and CaaS achieves absolute packet delivery, 42-45% lower latency, and three orders of magnitude lower jitter. We believe CaaS is a meaningful step towards the distribution, virtualization, and servitization of industrial control.

Flexible manufacturing is one of the core goals of Industry 4.0 and brings new challenges to current industrial control systems. Our detailed field study on auto glass industry revealed that existing production lines are laborious to reconfigure, difficult to upscale, and costly to upgrade during production switching. Such inflexibility arises from the tight coupling of devices, controllers, and control tasks. In this work, we propose a new architecture for industrial control systems named Control-as-a-Service (CaaS). CaaS transfers and distributes control tasks from dedicated controllers into Time-Sensitive Networking (TSN) switches. By combining control and transmission functions in switches, CaaS virtualizes the industrial TSN network to one Programmable Logic Controller (PLC). We propose a set of techniques that realize end-to-end determinism for in-network industrial control and a joint task and traffic scheduling algorithm. We evaluate the performance of CaaS on testbeds based on real-world networked control systems. The results show that the idea of CaaS is feasible and effective, and CaaS achieves absolute packet delivery, 42-45% lower latency, and three orders of magnitude lower jitter. We believe CaaS is a meaningful step towards the distribution, virtualization, and servitization of industrial control.

Future mobile devices are anticipated to perceive, understand and react to the world on their own by running multiple correlated deep neural networks locally on-device. Yet the complexity of these deep models needs to be trimmed down both within-model and cross-model to fit in mobile storage and memory. Previous studies squeeze the redundancy within a single model. In this work, we aim to reduce the redundancy across multiple models. We propose Multi-Task Zipping (MTZ), a framework to automatically merge correlated, pre-trained deep neural networks for cross-model compression. Central in MTZ is a layer-wise neuron sharing and incoming weight updating scheme that induces a minimal change in the error function. MTZ inherits information from each model and demands light retraining to re-boost the accuracy of individual tasks. MTZ supports typical network layers (fully-connected, convolutional and residual) and applies to inference tasks with different input domains. Evaluations show that MTZ can fully merge the hidden layers of two VGG-16 networks with a 3.18% increase in the test error averaged on ImageNet for object classification and CelebA for facial attribute classification, or share$39.61\%$parameters between the two networks with$<0.5\%$increase in the test errors. The number of iterations to retrain the combined network is at least$17.8\times$lower than that of training a single VGG-16 network. Moreover, MTZ can effectively merge nine residual networks for diverse inference tasks and models for different input domains. And with the model merged by MTZ, the latency to switch between these tasks on memory-constrained devices is reduced by$8.71{\times}$.

Accurate, real-time object detection on resource-constrained devices enables autonomous mobile vision applications such as traffic surveillance, situational awareness, and safety inspection, where it is crucial to detect both small and large objects in crowded scenes. Prior studies either perform object detection locally on-board or offload the task to the edge/cloud. Local object detection yields low accuracy on small objects since it operates on low-resolution videos to fit in mobile memory. Offloaded object detection incurs high latency due to uploading high-resolution videos to the edge/cloud. Rather than either pure local processing or offloading, we propose to detect large objects locally while offloading small object detection to the edge. The key challenge is to reduce the latency of small object detection. Accordingly, we develop EdgeDuet, the first edge-device collaborative framework for enhancing small object detection with tile-level parallelism. It optimizes the offloaded detection pipeline in tiles rather than the entire frame for high accuracy and low latency. Evaluations on drone vision datasets under LTE, WiFi 2.4GHz, WiFi 5GHz show that EdgeDuet outperforms local object detection in small object detection accuracy by 233.0%. It also improves the detection accuracy by 44.7% and latency by 34.2% over the state-of-the-art offloading schemes.

Time-Sensitive Networking (TSN) is the most promising network technology for Industry 4.0. A series of IEEE standards on TSN introduce deterministic transmission into standard Ethernet. Under the current paradigm, TSN can only schedule the deterministic transmission of time-triggered critical traffic (TCT), neglecting the other type of traffic in industrial cyber physical systems, i.e., event-triggered critical traffic (ECT). So in this work, we propose a new paradigm for TSN scheduling named E-TSN, which can provide deterministic transmission for both TCT and ECT. The three techniques of E-TSN, i.e., probabilistic stream, prioritized slot sharing, and prudent reservation, enable the deterministic transmission of ECT in TSN, and at the same time, protect TCT from the impacts of ECT. We also develop and make public a TSN evaluation toolkit to fill the gap in TSN study between algorithm design and experimental validation. The experiments show that E-TSN can reduce the latency and jitter of ECT by at least an order of magnitude compared to state-of-the-art methods. By enabling reliable and timely delivery of ECT in TSN for the first time, E-TSN can broaden the application scope of TSN in industry.

Accurate, real-time object detection on resource-constrained devices enables autonomous mobile vision applications such as traffic surveillance, situational awareness, and safety inspection, where it is crucial to detect both small and large objects in crowded scenes. Prior studies either perform object detection locally on-board or offload the task to the edge/cloud. Local object detection yields low accuracy on small objects since it operates on low-resolution videos to fit in mobile memory. Offloaded object detection incurs high latency due to uploading high-resolution videos to the edge/cloud. Rather than either pure local processing or offloading, we propose to detect large objects locally while offloading small object detection to the edge. The key challenge is to reduce the latency of small object detection. Accordingly, we develop EdgeDuet, the first edge-device collaborative framework for enhancing small object detection with tile-level parallelism. It optimizes the offloaded detection pipeline in tiles rather than the entire frame for high accuracy and low latency. Evaluations on drone vision datasets under LTE, WiFi 2.4GHz, WiFi 5GHz show that EdgeDuet outperforms local object detection in small object detection accuracy by 233.0%. It also improves the detection accuracy by 44.7% and latency by 34.2% over the state-of-the-art offloading schemes.