Embedded vision equipment of industrial robot for inline detection of product errors by clustering–classification algorithms

Abstract

The article deals with the design of embedded vision equipment of industrial robots for inline diagnosis of product error during manipulation process. The vision equipment can be attached to the end effector of robots or manipulators, and it provides an image snapshot of part surface before grasp, searches for error during manipulation, and separates products with error from the next operation of manufacturing. The new approach is a methodology based on machine teaching for the automated identification, localization, and diagnosis of systematic errors in products of high-volume production. To achieve this, we used two main data mining algorithms: clustering for accumulation of similar errors and classification methods for the prediction of any new error to proposed class. The presented methodology consists of three separate processing levels: image acquisition for fail parameterization, data clustering for categorizing errors to separate classes, and new pattern prediction with a proposed class model. We choose main representatives of clustering algorithms, for example, K-mean from quantization of vectors, fast library for approximate nearest neighbor from hierarchical clustering, and density-based spatial clustering of applications with noise from algorithm based on the density of the data. For machine learning, we selected six major algorithms of classification: support vector machines, normal Bayesian classifier, K-nearest neighbor, gradient boosted trees, random trees, and neural networks. The selected algorithms were compared for speed and reliability and tested on two platforms: desktop-based computer system and embedded system based on System on Chip (SoC) with vision equipment.

Effect of Superimposed Vibrations on Droplet Oscillation Modes in Prilling Process

“…The oscillation frequency was varied in the range of 200-800 Hz with an amplitude of 50 μm to study an effect of superimposed vibrations on droplet oscillation modes. The obtained images were analyzed using the Object Detection Method using Matlab software [40].…”

Possible implementations of the cloud-based data-centric IoT pilot platforms for smart cities and agriculture

“…It uses very little network traffic and is suitable for sensors that send their data to the "cloud". It requires connection to a type of dispatcher (Broker), who receives a subscriber of a particular sensor / device on a topic [5][16] [25]. The Message Queue Telemetry Transport Protocol (MQTT) was developed by Andy StanfordClark (IBM) and Arlen Nipper (Eurotech) in 1999 to track the pipeline in the middle of the desert.…”

The Digitization of Quality Control Operations with Cloud Platform Computing Technologies

“…Industrial vision systems are computer-based systems where software performs tasks for acquiring, processing, analyzing, and understanding digital images usually aimed at industrial quality assurance, defect detection, part recognition, etc. Research articles concerning industrial vision systems describe, for example, an automatic surface detection (Zhou et al 2019), pre-inspection of steel frames (Martinez et al 2019), embedded vision systems (Zidek et al 2016), and stereo vision sensing (OrRiordan et al 2018). Additional related research aspects of RFID system were presented in works focused on the security of tags (Han et al 2019), detection of missing tags (Lee et al 2019), and new searching protocol (Liu et al 2019).…”