Machine Learning for Auto Optimization What is Machine Learning? - - PowerPoint PPT Presentation

Machine Learning for Auto Optimization

What is Machine Learning? Definition: “Machine learning refers to any system where the performance of a machine in performing a task improves by gaining more experience in performing that task”.

• Experience refers to the data that we fed in to the algorithm and improvements refers to it output

which is considered as an action.

• ML is intelligence acquired by a machine, which is similar to human natural intelligence.
• ML use existing data to forecast future behaviors, outcomes, and trends.
• ML involves using statistical / mathematical techniques.

Examples of Machine Learning A computer program is said to learn from experience E with respect to task T and performance gauge P.

• Optical Character Recognition: categorize images of hand written characters by the letters

represented.

• Face detection: Find faces in Image.
• Spam Filtering: identify email messages as spam or non spam.

ML Algorithm Performance measuring(P) Experiences (E) Task (T) Traffic pattern(T) Future Traffic pattern(P) Historic Traffic pattern(E)

Applying Machine Learning to CNC Machines

• Performance Improvement using Machine

Learning:

• Thermal Displacement Compensation
• Automatic Servo Tuning
• Adaptive Control for optimizing cycle time.
• Learning control for achieving high performance

machining.

• Inertia Estimation, for higher acceleration to

reduce cycle time.

• Smart Program Analysis – Acc/dec decided

dynamically

• Preventive Maintenance using Machine

Learning:

• Prediction of Failures
• Data Analysis using AI - Pattern Analysis/

Waveform Analysis

• Minimizing Downtime using AI

Thermal Displacement Compensation Conventional Method  It is not easy to derive the relationship between temperature and displacement necessary for thermal displacement compensation

：Temperature sensor

：Displacement sensor

Temp. Disp.

Analysis, Formulation Heat transfer analysis Thermal fluid analysis. etc.

Data Collection Software development

Thermal Displacement Compensation Using machine learning  Machine learning can derive the relations from the data of temperature and displacement and can create thermal displacement model.

Model Development Software

：Temperature sensor ：Displacement sensor

Machine Learni arning TDC Model

Thermal Displacement Compensation

Thermal Displacement Compensation option

Model development tool

Learning Data Data Collection Software Temp. Disp. Temp. Comp.

Automatic Servo Tuning

• Auto-tuning of servo gain and Acc/Dec time constant according to target work

piece

• Useful for machining optimization

Ethernet Machine tools

Workpiece1

Collect Collect Restore

Workpiece 2 Workpiece 1 SERVO Tuning Data 1 SERVO Tuning Data 2 SERVO Tuning Data 1

:

Manage

Workpiece2 SERVO Tuning Data 2 Workpiece1 SERVO Tuning Data 1

Inertia Estimation, For higher acceleration to Reduce Cycle Time

• Can automatically estimates the inertia when Job changes.
• Can achieve optimum positioning time.

Adaptive Control for Optimizing Cycle Time

• Automatic Feed rate control according to spindle load and temperature.
• Controlling feed rate according to spindle load strikes a good balance between

shorted cycle time and longer life time of cutting tools.

Adaptive Control for Optimizing Cycle Time

Learning Control for Achieving high performance machining Servo learning Control

• Suppress periodic machining disturbance.

Learning Control for Achieving high performance machining Servo learning Oscillation

• Avoid chip Entanglement by oscillation cutting for chip shredding using servo

learning.

• Contribution to productivity improvement by continuous operation.
• Reduction of production costs by elimination of chip removal system.

Smart Program Analysis-Acc/Dec decided dynamically

• Artificial Intelligence Contour Control Function for reading small segments of program in

advance and will create smooth profile.

Prediction of Failures- AI Spindle Monitor

• Anomaly monitoring of spindle by machine learning.
• Can predict the spindle failure in advance.

Model creation at normal state Calculation of Anomaly score Acquisition of servo data

Data Analysis Using AI- Pattern Analysis/Waveform Analysis

• Monitor the servo and spindle loads and establish pattern(Signature) for the

component.

• Collect servo data with high speed

sampling (1ms) and to store with file format

• Displays collected data for

analysis. Collection of various sensors data and servo data

• Collect data from various sensors

(temperature, shock etc.) via CNC by using i/o units. . . .

Database Operation Management software

Analog interface module External sensor

Servo data

VIEWER software

Servo data Motor speed Machine Acc.

MULTI SENSOR I/O UNIT Temperature sensor Shock sensor

Sensor data

Applications

Minimizing Down time using AI

• Manages diagnosis information of Trouble Diagnosis and Machine Alarm

Diagnosis with final solutions when alarm occurs.

• When newly alarm occurs, indicate solution from similarly diagnosis information

Normal Trouble Diagnosis AI Trouble Diagnosis

• Operator implement diagnosis according to

CNC guidance/Manual.

• Operator needs to diagnose when multiple

estimation causes finally to be left

• AI indicate higher probability solution from past history data.
• Automatic judgment from countermeasure / treatment information in case of

multiple estimation causes remained.

Rapidly restoration at trouble Collect Alarm!

Actually Measures/ Treatments

Add Indicate

Applying Machine Learning for Robotic Automation. Faster Bin Picking application:

• Robots automatically learns the picking sequence of work piece.
• Drastically reduces the time for manual setting and tuning.

AI Bin picking Application

Applying Machine Learning for Robotic Automation. Learning Vibration Control: Learning robot realizes high speed smooth motion with suppression of vibration by LVC (Learning Vibration Control). Learning robot merit This function has

• vercome vibration

issues of high speed motion, which has not be used before.

W/ LVC W/O LVC

Accelerometer

Cycle time can be reduced by high speed motion. (i.e. realization of higher performance for each )

Vibration Suppressed!!

Learning Control + Sensor Technology

Prediction of Failures- Mechanical Failures

• To eliminate unplanned downtime

. Maintenance health Process Health System Health

Grease replacement Battery replacement Greasing to the balancer bush

Vision detection result Welding current monitor Servo gun status monitor Operational status

Memory usage Alarm information

Increasing vibration of J2!

Reducer to be exchanged next weekend.

proceed production proceed production proceed production proceed production

Mechanical Health

Replace grease !

Alarms

Machine Learning on Standalone Vs network of Systems Stand alone Machine with networking

• Learning with experience is confined to
• ne machine.
• Learning will be vast since all machines

will be sharing there data and solution can be immediately found.

Server with ML

Machine learning With IOT IoT IoT- Conn nnects cts Thing ngs s – “Internet of Things”

• IOT provides a platform on which number of devices are

connected and pushing down data in a centralized system.

• IoT devices follow these five basic steps: measuring,

sending, storing, analyzing, acting.

• The collected datasets are fed into Machine learning

algorithms to take active decisions.

Cloud Computing ON Premises ON CLOUD

2016 2017 2018 2016 2017 2018

• In IOT System,

to save huge amount of data, known as Big Data, stack of storage devices are required.

• IOT data will be increasing exponentially &

hence will require frequent hardware up gradation.

• To Run Machine learning/AI algorithms, high

computation power processors are required and single processor is not sufficient.

Advantages of Cloud Computing

Flexibility If your needs increase it’s easy to scale up your cloud capacity, drawing on the service’s remote servers. Disaster recovery Businesses of all sizes should be investing in robust disaster recovery,. Automatic software updates Suppliers take care of servers for you and roll out regular software updates. Capital-expenditure Free Cloud computing cuts out the high cost of hardware. Work from anywhere With cloud computing, if you’ve got an internet connection you can be at work.

FOG Computing

• FOG Computing is an intermediate

layer between device and Cloud.

Cloud FOG (T3 Time for processing) IOT Devices (T1 time for data generation) T2 Sec T4 Sec On- Premises Non-Critical data sent directly After processing data is saved in cloud Data Segment Non- Critical Data Critical Data

IOT/Cloud computing with ML

• The only way to analyze the data generated by the IoT is with machine learning/AI.

FANUC Solutions for Machine learning & IOT.

AI and Cloud computing Data collection and Monitoring (IOT)

Visualization

MT-LINK i

Diagnosis

Notification Host system software Communication interface Data collection

Connecting Collecting

Communicating

Smart/AI Features

FANUC Intelligent Edge Link & Drive system

FANUC MT-LINK i

• AI Thermal displacement

compensation

• AI Servo Tuning.
• AI spindle monitoring.
• AI contour control(AICC).
• AI Bin Picking
• Smart Adaptive Control
• Smart Feed axis Acc/Dec
• Servo learning Control.
• Zero Down Time(ZDT)
• Learning Vibration control

Thank You for your Kind attention