Method and system for recommending tool configurations in machining

Patent metadata

Field	Value
Publication number	US-12165083-B2
Application number	US-202017111097-A
Country	US
Kind code	B2
Filing date	Dec 3, 2020
Priority date	Dec 4, 2019
Publication date	Dec 10, 2024
Grant date	Dec 10, 2024

Abstract

This disclosure relates generally to recommending tool configurations in machining. The machining tool configuration selection involves the selection of several tool specification parameters concerning the material, geometry and composition of the machining tool. The state-of-the-art methods uses a rule and knowledge-based system to select tool configuration, however these methods do not recommend tool configurations which satisfy customer requirement. Embodiments of the present disclosure uses a hierarchical model which is trained to predict acceptable tool specification parameters for a given requirement by learning the patterns from past tool selection data. Further a probabilistic approach is used to predict the top set of recommendations of tool configurations with a probability score for each prediction. The disclosed method is used for recommending tool configurations in a cylindrical grinding wheel process.

First claim

What is claimed is:

1. A processor implemented method for recommending tool configurations in machining ( 200 ) using a machine learning approach and a probability model, the method comprising:
receiving, via one or more hardware processors, customer requirement data as input for recommending a set of tool configurations in machining, wherein customer requirement data for a cylindrical grinding wheel process includes material removal rate (MRR), workpiece material, workpiece hardness, surface roughness, cylindricity, dressing parameters, machine specifications, workpiece dimensions, and grinding parameters, wherein each tool configuration of the set of tool configurations comprises a set of tool configuration parameters ( 202 ), wherein the set of tool configuration parameters comprise machining tool parameters;
extracting, via the one or more hardware processors, a set of model input parameters from the customer requirement data based on a predefined machining process ( 204 ) comprising one or more of the cylindrical grinding wheel process, a turning process, or an end milling process, wherein the set of model input parameters comprises one or more of (i) work material parameters including workpiece geometry parameters, workpiece material and material hardness, (ii) machining parameters including a spindle speed, a feed rate and a depth of cut, (iii) quality requirement parameters including a dimensional accuracy and a surface finish, (iv) tool life requirement parameters, or (v) machining environment parameters;
obtaining, via the one or more hardware processors, a probability score for each tool configuration parameter of the set of tool configuration parameters from the set of model input parameters using a hierarchical model, wherein the hierarchical model is built considering inter relationships between tool configuration parameters and selection sequence of the tool configuration parameters and pre-trained ( 206 ) based on a sequence of selection of the set of tool configuration parameters and corresponds to a selection hierarchy of a material selection followed by a machining tool parameter selection,

wherein training of the hierarchical model comprises:
receiving a set of past tool selection data as a first set of training data for training the hierarchical model ( 302 );
identifying a set of model parameters from the set of past tool selection data based on the predefined machining process, wherein the set of model parameters comprises the set of model input parameters and the set of tool configurations ( 304 );
augmenting the first set of training data to obtain a second set of training data based on a number of the first set of training data ( 306 ), wherein an augmented data is created by varying parameters from the received first set of training data, wherein the first set of training data comprises continuous and discrete parameters, the continuous parameters are augmented using a Synthetic Minority Over-sampling Technique (SMOTE), SMOTE-NC (Nominal and Continuous); and
training the hierarchical model to predict each of the tool configuration parameters corresponding to the set of model input parameters using one of (i) the first set of training data or (ii) the second set of training data ( 308 ), and along with the training, hyper parameter tuning is performed for obtaining performance in terms of accuracy, wherein a first model is trained to predict a first tool configuration parameter thereby the hierarchical model learns insights from the data with a decrease in a number of varied classes for the predefined machining process, and a second model is trained to predict a second tool configuration parameter,
wherein the second model is trained by feeding an input data of the first model input parameters and a predicted class of the first model, wherein a number of models in the hierarchical model is dependent on a number of tool configuration parameters to be recommended in the predefined machining process;
determining, via the one or more hardware processors, a joint probability for each tool configuration of the set of tool configurations using the probability score for each tool configuration parameter ( 208 ), wherein an abrasive name model, a bond name model, a Grit name model, a wheel structure name model and a wheel grade name model predicts the probability score for the tool specification parameters abrasive name, bond name, grit name, structure name and wheel grade, wherein model creation is performed using a semi deep neural network architecture wherein for each model a 5-layer neural network is trained and tested on the augmented data and after training, the models are saved for future use, so that they are called for predictions on new data points; and
recommending, via the one or more hardware processors, a predefined number of tool configurations out of the set of tool configurations, that are new combinations of tool specification parameters, based on a predefined criterion of the joint probability of each tool configuration ( 210 ).

2. The processor implemented method as claimed in claim 1 , wherein augmenting of the first set of training data is performed if the number of the first set of training data is less than a predefined threshold, wherein the predefined threshold is dependent on number of the set of model parameters.

3. The processor implemented method as claimed in claim 1 , wherein a set of past tool selection data comprises past customer requirement data and corresponding selected tool configurations.

4. The processor implemented method as claimed in claim 1 , wherein the predefined criteria of the joint probability of each tool configuration is a highest joint probability defined by a domain expert.

5. The processor implemented method as claimed in claim 1 , wherein in the material selection, an abrasive material is selected first, followed by selecting a bonding material based on the selected abrasive material and depends on the predefined machining process considered and a type of tool selected,
wherein the machining tool parameter selection includes tool geometry parameters and tool auxiliary parameters,
wherein in the turning process, the tool geometry parameters are cutting edge angles, and a nose radius, wherein in the cylindrical grinding wheel process, the tool geometry parameter is abrasive grit size, wherein the tool auxiliary parameters in the cylindrical grinding wheel process are wheel grade and wheel structure, and the tool auxiliary parameters depend depends on a specific tool and an application.

6. A system ( 100 ), comprising:
a memory ( 104 ) storing instructions;
one or more communication interfaces ( 106 ); and
one or more hardware processors ( 102 ) coupled to the memory ( 104 ) via the one or more communication interfaces ( 106 ), wherein the one or more hardware processors ( 102 ) are configured by the instructions to:
receive customer requirement data as input for recommending a set of tool configurations in machining, wherein customer requirement data for a cylindrical grinding wheel process includes material removal rate (MRR), workpiece material, workpiece hardness, surface roughness, cylindricity, dressing parameters, machine specifications, workpiece dimensions, and grinding parameters, wherein each tool configuration of the set of tool configurations comprises a set of tool configuration parameters, wherein the set of tool configuration parameters comprise machining tool parameters;
extract a set of model input parameters from the customer requirement data based on a predefined machining process comprising one or more of the cylindrical grinding wheel process, a turning process, or an end milling process, wherein the set of model input parameters comprises one or more of (i) work material parameters including workpi

Assignees

• Tata Consultancy Services Ltd

Inventors

• Sharma Sunil
• Muhammed Bilal
• Pusuluri Srimannarayana
• Basavarsu Purushottham Gautham
• Das Prasenjit

Classifications

• G06N3/0499

  Feedforward networks · CPC title

• G06N3/0985

  Hyperparameter optimisation; Meta-learning; Learning-to-learn · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06N3/047

  Probabilistic or stochastic networks · CPC title

• G06N3/08

  Learning methods · CPC title