Dataflow Domain Research Articles

A Data-Flow Oriented Deep Ensemble Learning Method for Real-Time Surface Defect Inspection

Real-time surface defect inspection plays an important role in the quality control of automated production. For the surface defect inspection, flowing data are a common data form in the automated pipeline production. Although flowing data provide rich information for surface defect inspection, there are still a lot of dynamic distribution challenges caused by the flowing data, such as the domain shift phenomenon and imbalanced training data. However, many existing industrial inspection solutions still use static strategies. To determine the dynamic distribution influence on the data flow domain, this article proposes a new deep ensemble learning method with domain fluctuation adaptation. Specifically, a new distribution discrepancy identifier based on estimation of the data set distribution and data characteristic is proposed. It utilizes advantages of both the deep convolutional neural network (CNN) and the shallow feature-based learning method to achieve higher robustness and fine-grained detection in streaming data scenes. In order to validate the proposed method, an inspection bench test system, as a part of a real industrial surface mount technology production line, is designed and fabricated. The proposed inference model is successfully applied to an embedded terminal with a hybrid and heterogeneous computing architecture. At last, the method is validated on the data collected from the manufacturer. The result suggests that the proposed method possesses a competitive mean average precision (mAP) rate with good adaptation and robustness in industrial streaming data scenes.

UniTi: Unified Composition and Time for Multi-domain Model-based Design

To apply model-based design to embedded systems that interface with the physical world, including simulation and verification, current tools fall short. They must provide mathematical (model) definitions that stay close to the specification of the system. They must allow multiple domains, such as the continuous-time, discrete-time and dataflow domain, in a single model including well-defined interaction. They must support model transformations for refining a model during development. And most importantly, they must accurately include and simulate different notions of time in the model. UniTi is a model-based design flow and modelling and simulation environment that delivers on all these aspects. It is based on components that are signal transformations, and therefore mathematical functions. However, in each domain the representation of a signal differs. As components have the same structure in each domain, we can use unified composition operators to represent multiple domains in a single model. Furthermore, this composition provides a unified perspective on time in the domains, even though we differentiate between different notions of time. Time becomes a local property of the model, allowing us to represent and simulate time transformations such as time delays exactly without losing efficiency. Finally, model transformations are defined for such components, which are used for refining and developing the model and which are guided by the design steps in the design flow. We will formally define the domains, composition operators and transformations of UniTi and verify the approach with a case study on a phased array beamforming system.

Design and evaluation of a compiler for embedded stream programs

Applications that combine live data streams with embedded, parallel, and distributed processing are becoming more commonplace. WaveScript is a domain-specific language that brings high-level, type-safe, garbage-collected programming to these domains. This is made possible by three primary implementation techniques, each of which leverages characteristics of the streaming domain. First, we employ a novel evaluation strategy that uses a combination of interpretation and reification to partially evaluate programs into stream dataflow graphs. Second, we use profile-driven compilation to enable many optimizations that are normally only available in the synchronous (rather than asynchronous) dataflow domain. Finally, we incorporate an extensible system for rewrite rules to capture algebraic properties in specific domains (such as signal processing). We have used our language to build and deploy a sensornetwork for the acoustic localization of wild animals, in particular, the Yellow-Bellied marmot. We evaluate WaveScript's performance on this application, showing that it yields good performance on both embedded and desktop-class machines, including distributed execution and substantial parallel speedups. Our language allowed us to implement the application rapidly, while outperforming a previous C implementation by over 35%, using fewer than half the lines of code. We evaluate the contribution of our optimizations to this success.

Emulation of a parallel codeblock dataflow processor

The paper describes an experimental multiprocessor for research experiments in the coarse-grain dataflow domain where a combination of dataflow concepts and conventional von Neumann-style processing aims at providing a powerful computation concept suitable for technical/scientific and real-time system applications.