You are on page 1of 20

Data Mining for

Manufacturing
Analytics

Helps in achieving
Decision Making
Analytics

Objectives

Data Mining
Data

What data ?

Which data mining tool ?

Which decision making tool ?

Objectives
Example:
Finance
Customer

What data ?

Process

Which data mining tool ?

Which decision making tool ?

Dimension
Finance
Customers

KPI

Productivity

Become industry cost leader % reduction in cost/unit

Growth
Quality
Timeliness

Increase market share


Zero defects
On time delivery

Reduce inspections
Reduce time to produce
Improve reliability

Market share
% good quality first pass
% on time deliveries
% orders delivered to
assembly
% suppliers with ISO 9000
Cycle time
Warranty costs

Post-Sales
services

Improve response to
customer inquiries

% querries satisfied on first


pass

Risk

Reduce inventory
obsolescence

Inventory turnover

Reduce customer backlog

% backlogged orders

Suppliers
Products
Processes

Objectives

Human capital
Learning and Information
capital
Growth
Organizational
capital

Integrate into production

Develop quality improvement % trained in SPC or Six


skills
sigma
Provide technology to
improve processes

% customers who can track


orders

Create innovative culture

% NPD projects per year

Current
status

Goal for
2017

Objectives

Shop-floor Issues

What data ?

Which data mining tool ?

Which decision making tool ?

Workpiece Measurements
Increased
Bearing
Clearances

Workpiece
measurements

Diagnosis
Diagnosis

Integrating workpiece measurements with machine monitoring


parameters will increase early detection probability

very smooth
smooth
rough
very rough
failure

Vibration
Parameter
0 to 0.1
0.1 to 0.15
0.15 to 0.22
0.22 to 0.37
> 0.37

Inspection interval = 30 days


0.0 to
0.1 to
0.15 to
0.10
0.15
0.22
0.575
0.224
0.145
0.249
0.331
0.377

0.22 to
0.37
0.045
0.136
0.229
0.242

> 0.37
0.011
0.284
0.394
0.758
1.000

Proportional Hazards Model


The most commonly used model in CBM is the Proportional Hazards Model
(PHM) with covariates for describing the current state

[ a1 x1 a2 x2 ........ an xn ]

h(t ) h0 (t ).e

This model is useful for continuous monitoring


BUT
Cannot predict in to future

[ a1 x1 ( t ) a2 x2 ( t ) ........ an xn ( t )]

h(t ) h0 (t ).e

Process Measurements

Power
Consumption
measurements

Time

Courtesy: MGTL

Workpiece Measurements
Increased
Bearing
Clearances

Control
Charts

Workpiece
measurements

Diagnosis
Diagnosis

Integrating workpiece measurements with machine monitoring


parameters will increase early detection probability

Improper coolant concentration and pH levels


can not only reduce tool life and cutting
accuracy, but can also lead to severe damage
to the machine tool and its components. The
coolant monitoring system facilitates proactive
maintenance to better protect your investment.

Maintenance of appropriate cutting fluid at the


proper concentration means higher feed rates,
deeper cuts, better accuracy, reduced power
consumption, and longer tool life.

www.okuma.com

Collision
Sensors

Vibration
Sensors

Accessory
Condition
Monitoring

Process
Sensors

Temp
Sensors

IoT as an Enabler
Increased
Bearing
Clearances

Workpiece
measurements

Data
DataStorage
Storage

IoT as an Enabler

The
The Internet
Internet of
of Things
Things (IoT)
(IoT) is
is the
the interconnection
interconnection of
of
uniquely
uniquely identifiable
identifiable embedded
embedded computing
computing devices
devices
within
within the
the existing
existing Internet
Internet infrastructure.
infrastructure.

Sub System
Bearing
Algorithm (URL)

Distributed

Parameter 1

Analytics

Is stored in

Parameter 2
Threshold value
Value for good condition
Knowledge
Base

Health Status

Algorithms
Parameters
Rule base
Threshold values
Knowledge
Base

Baselines

Residual
Residual Life
Life prediction
prediction

Analytics Tool-Box
Important dimensions of such a tool box are considered below to visualize
its possible scope.
Application Area
Quality Analytics
Analysis type
Online, Off-line
Domain
Time, Frequency (all of the following may not fit in each domain)
Variable type
Continuous, Discrete
Parameter to be tracked
Univariate, Multivariate
Data type
Series: Uncorrelated, Auto-correlated
Events: Temporal, Non-Temporal

Analytics Tool-Box
Data Behaviour
Series: In-control, Freaks, Repeats, Cycles, Shifts, Drifts, Mixed patterns,
Abnormal fluctuations, Inconsistent variance,
Events: Random, Independent, Dependent, Repetitive
Tasks
Classification: Determining the category to which a data-set belongs
Assessment: The act of making a judgment about something. For eg., Is
the process in control?
Diagnosis: Analysis of a system state
Monitoring: Watching out for discrepancies.
Prediction: When is an event likely to occur? How will the process behave
in the future?
Typical Analysis Path
Monitoring provides inputs for Assessment which helps in correct Diagnosis
leading to correct Classification and more accurate Prediction

Analytics Tool-Box
Method/Tool choosing criteria
RISE: Robust, Intuitive, Simple to use, Effective
Approach for building the software:
Identify possible scenarios as a combination of above. For eg., Online
monitoring of a auto-correlated continuous variable representing a
product quality characteristic with a possibility of drift in the mean
values
Identify/Create/Simulate data-sets
Identify tools
Evaluate them on the RISE criteria
Apply tools
Analyse/compare their performance

You might also like