Machine Design to Machine Learning

Source: Deep Learning on Medium

Machine Design to Machine Learning

Machine Learning is widely used in the software industry for a variety of reasons. What about the hardware industry who has not kept the data (mechanical industries used to enter the data in the logbook, there are still small scale industries keeping their data in the logbook).

In mechanical engineering, ML techniques are used in various applications such as physical behavior estimation, design, recognition, reverse engineering or material sciences. In those applications, classifiers are often used to estimate one or several output parameters of different natures (e.g. geometrical, statistical, physical), or even to classify shapes or 3D points sets. For sizing and shape design, classifiers are often estimating global geometric parameters of the model. The estimation of a physical quantity like a load, stress, pressure or temperature, remove the need to solve complex equations. Statistical parameters do not give directly the physical quantity. However, they offer the opportunity to estimate for example a standard deviation, a mean, a trend or a physical effect probability. The classification of shapes or digitized 3D points sets is used for model recognition and reuse.

Machine Learning application in Mechanical field

  1. Predictive Maintainance — Understand the health of the machine by taking the data from the machines through different sensors and predict the behavior on the basis of failure criteria.
  2. 3D model classification — Deep convolution neural networks can be successfully applied to classify the 3D CAD models into different segments. This technique can be used to quickly retrieve the 3D data from the database.
Deep Convolutional Neural Network (CNN) performing an object recognition task.

3. Time series analysis — This technique is used for Demand and sales forecasting, Inventory and stock scheduling Planning, dispatching and scheduling.

4. Object Recognition — The challenge of object recognition can be summarised as follows:
“A vision system which makes use of an object model is referred to
as a model-based vision system, and the general problem of identifying
the desired object is referred to as object recognition. While there is no
single definition of the object recognition problem, the objective is to
identify a desired object in the scene and to determine its exact location
and orientation.”

Fig. given below presents five tasks in which the research area of Object Re-
cognition can be sub-divided. Each of the following tasks aims to address slightly different objectives with an increasing difficulty yet essential to overcome in order to specify and build a system able to perform cross-comparisons between 3D and 2D data.

Object recognition can be subdivided into five different tasks with increasing difficulty.

5. Feature Recognition — Automated machining feature recognition, a sub-discipline of solid modeling, has been an active area fo the last three decades and is a critical component in a digital manufacturing information from CAD models. Machining features (holes, pockets, slots) can be easily recognized by using 3D convolutional neural networks.

The architecture of the CNN network trained to recognize machining features on 3D CAD models

It has been observed that deep learning algorithms and routines may present many desirable advantages and assets to solve challenges inherent to the industry of the future. There are more areas in the mechanical domain where we can use Machine Learning to learn from the existing data.

References —

  1. Jonathan Dekhtiar, Alexandre Durupt, Matthieu Bricogne, Benoit Eynard,
    Harvey Rowson, Dimitris Kiritsis. ”Deep learning for big data applications in CAD and PLM — Research review, opportunities and case study”, Computers in Industry 100 (2018) 227–243
  2. ZhiboZhang, PrakharJaiswal, RahulRai, ”FeatureNet: Machining feature
    recognition based on 3D Convolution Neural Network”, Computer-Aided Design Volume 101, August 2018, Pages 12–22