Download as pptx, pdf, or txt
Download as pptx, pdf, or txt
You are on page 1of 44

PREDICTIVE MAINTENANCE OF CNC MACHINE

USING MACHINE LEARNING

GUIDE:
Dr. A. JOY CHRISTY Team Members:
Assistant Professor II (SoC) P.M.Dinesh - 124161011
V.Easwar - 124161012
Siva Sankar - 124161037
S.NO TOPICS PG.NO

1 Base Paper Details 3


2 Problem Statement 4
3 Abstract 5
4 Introduction 6
5 Literature Survey 7
6 Workflow 11
7 Work carried out 12
8 Dataset details 13
9 Program 14
10 Insights 15
11 Feature scaling: 25
12 Logistic Regression 28
13 Training & accuracy score of target 30
S.NO TOPICS PG.NO
14 Random Forest 32
15 Training & accuracy score of Failure type 34

14 Results 36
15 Sample prediction 38
16 Comparison 39
17 Conclusion 40
18 Reference 41
BASE PAPER DETAILS:

BASE PAPER NAME:


Explainable Artificial Intelligence for Predictive Maintenance
Application

AUTHOR: Stephan Matzka

YEAR: 2020

LINK: https://ieeexplore.ieee.org/document/9253083
PROBLEM STATEMENT

A Machine like a CNC only receives maintenance after a set number of runs or when an issue first
arises. If that's the case, it could result in expensive issues and have a negative impact on safety. This will
result in a high rate of machine downtime in that business, which lowers industrial efficiency and negatively
impacts the company's ability to generate revenue. Predicting the maintenance required beforehand
accurately is the most challenging thing till today

Solution:

In this project, our main goal is to use Classification algorithms such as Decision Tree, Random
Forest, and SVM to predict when the CNC machine requires maintenance
ABSTRACT
❖ The project is about to generate an efficient machine learning model for the predictive maintenance
of CNC machines.

❖ Any breakdown in the CNC , due to electrical, environmental, or tooling failures, may require a
combination of mechanical and programming expertise to get it fixed, so catastrophic breakdowns
can be easily avoided with predictive maintenance.

❖ This project uses a supervised learning model to learn and predict the labels for new data based on
input features.

❖ Classification algorithms such as Decision Tree, Random Forest, and SVM are used to predict
discrete outcomes.

❖ The main outcome of the project is to determine if a machine could fail and if so, determine the
nature of the failure.

❖ Prediction accuracy using the same SL model can be completely different for different training sets.
INTRODUCTION
❖ Using a process called predictive maintenance, it is possible to assess the state of a piece of
machinery and predict when maintenance needs to be done.

❖ This is done by collecting data from sensors and analyzing it to detect any potential problems.

❖ It can be used to detect patterns in the data that indicate a problem is developing, such as vibration or
temperature changes. This allows for preventive maintenance to be scheduled before the issue
becomes a more serious problem.

❖ Predictive maintenance offers many benefits, including improved safety, reduced downtime, and cost
savings.

❖ It can help to reduce the risk of accidents or injuries by identifying potential safety issues before they
become a problem.
LITERATURE SURVEY
PAPER AUTHOR YEAR MERITS DEMERITS ACHIEVEMENT

"Explainable D. Gunning 2017 High level of Dynamics of Create AI systems


artificial accuracy. data and whose learned
intelligence decisions( does models and decisions
(xai)." not guarantee can be understood
the correct and appropriately
decision). trusted by end users.

“Interpretable C. Molnar 2020 Help increase Lack of Learnt methods to


machine trust in machine universal make any machine
Learning” learning . interpretability. model interpretable.
“Using Process S. Matzka 2018 Increased Limited Machine learning
Quality productivity. applicability. concepts to predict
Prediction to process quality early in
increase the process is proposed
Resource
Efficiency in
Manufacturing
Processes”

“AI4I 2020 S. Matzka 2020 Consists of Simulated data. provide a synthetic


Predictive 10,000 instances dataset that reflects real
Maintenance and predictive maintenance
Dataset” 10 Attributes. encountered in industry
to the best of our
knowledge.
“Estimates of Mathworks 2020 Easy to Limited to Used to determine
predictor implement classification which features or
importance for tasks predictors in a
classification dataset are most
ensemble of important in making
decision trees” predictions.

"Why should I Ribeiro, 2016 Multiple The framework Deciding if one


trust you?: Marco Tulio, explanation does not provide should trust a
Explaining the Singh, technique. guidance on how prediction, choosing
predictions of Sameer, to handle between models,
any classifier." Guestrin, situations where improving an
Carlos the explanations untrustworthy
contradict the classifier, and
user's prior identifying why a
beliefs or classifier should not
expectations. be trusted.
WORKFLOW

GOAL OBJECTIVE TIMELINE


Collecting a dataset from source,
I Analysing the dataset , Data Month 1
preprocessing

Algorithm choosing ,splitting the


II Month 2
data into Train and test dataset ,
Model training

III Evaluation , Predicted diagnosis , Month 3


Deploying the model in an interface
and Documentation
WORK CARRIED OUT
DATASET DETAILS
Since real predictive maintenance datasets are generally difficult to obtain, we’ve used a synthetic
dataset that reflects real predictive maintenance data encountered in industry. The dataset consists of
10,000 data points stored as rows with 10 features in columns.

Some key features are:

1. Air Temperature 6. Tool wear

2. Process Temperature 7. Target

3. Product type 8. Failure Type

4. Torque

5. Rotational Speed

Reference:

www.explorate.ai/dataset/predictiveMaintenanceDataset.csv
FAILURE TYPES COUNT BASED ON AIR TEMPERATURE
FAILURE TYPES COUNT BASED ON PROCESS
TEMPERATURE
FAILURE TYPES COUNT BASED ON TEMP DIFF

➔ Heat Dissipation Failures are more dependent on temperature difference than either the air or process
temperature.
ROTATIONAL SPEED[RPM] VS TORQUE[Nm]

➢ The machine fails either for the lowest or largest values of torque and rotational speed. This is easily spotted in
the graph since the orange dots are far apart for those features. So, there is a range for normal conditions in
which the machines operate, and above or under this range, they tend to fail.
➢ Heat dissipation
failure only occured
when the RPM was
below 1500 and it
seems to be
independent of tool
wear
➢ More clearly to say,
it occurs at lower
RPM and higher
torque when the
temperature
difference is less,
meaning it occurs
more frequently
when the machine is
starting

➢ Power failures either occur at high torque low rpm or high rpm low torque
ROTATIONAL SPEED[rpm] VS TOOL WEAR[min]
TORQUE[Nm] VS TOOL WEAR[Min]

➢ Tool
wearing
failure
occurs
mostly
when the
tool wear
min is high,
it is
independent
of rpm and
torque.
Between
torques 16
Nm and 41
Nm all
failures are
tool wear.
ROTATIONAL SPEED[Rpm] VS TEMP DIFF
TORQUE[Nm] VS TEMP_DIFF

➢ Over strain failure happens at higher torque and lower RPM but unlike heat dissipation failure, they are
independent of temperature difference but they need high tool wear min similar to tool wearing failure
Program:
https://drive.google.com/drive/folders/1_tF84D71LT_jOASt0F4uogi_ChxQ0UYt?usp
=sharing
HEATMAP:
FEATURE SCALING:
1) ROBUST SCALER:
2) MINMAX SCALER 3)ORDINAL ENCODER
LOGISTIC REGRESSION:

Logistic regression is a ML classification algorithm that is used to predict the probability of


certain classes based on some dependent variables. The output of logistic regression is always between
(0, and 1), which is suitable for a binary classification task. The higher the value, the higher the
probability that the current sample is classified as class=1, and vice versa.
Logistic Regression can be expressed as,
Log(p/1-p) = β0 + β1*x

β0 is the bias or intercept term and β1 is the coefficient for the input value (x)
A sigmoid curve is used to represent the relationship between the input variables and the
probability of a certain outcome. The sigmoid curve in logistic regression is shaped like an "S" and is
called the logistic function. It takes the input and transforms it into a value between 0 and 1. At the low
end of the curve, the probability is close to 0.
In the middle of the curve, there is a rapid increase in the probability Finally, at the high end of the
curve, the probability approaches 1.
TRAINING & ACCURACY SCORE OF TARGET:
CONFUSION MATRIX
RANDOM FOREST ALGORITHM

Random forest is an ensemble classifier, i.e. a combining classifier that uses and
combines many decision tree classifiers. The reason for using a large number of trees in
random forest is to train the trees enough such that contribution from each feature comes in a
number of models. After the random forest is generated by combining the trees, a majority
vote is applied to combine the output of the different trees.
The working of Random Forest:
 Select random samples from a given training set.
 This algorithm constructs the Decision Tree from Training data.
 Average voting would be done
 Then, the predicted results will be achieved.
TRAINING & ACCURACY SCORE OF FAILURE TYPE:
CONFUSION MATRIX
RESULTS
OPERATIONALIZE:

● We have found a model that is good enough at addressing the problem.

● It is all the way up to integrating the model into an existing enterprise application.

ALGORITHM IMPLEMENTATION:

● To increase the accuracy score, we trained our data using logistic regression and random
forest.

● When compared to other algorithm like Decision tree, these two provided us with a greater
accuracy rating.
● In training we achieved an accuracy score of the model for the target is 97%, in testing
we achieved an accuracy score of 97.6%.

● In training we achieved an accuracy score of the model for the Failure type
99.99%, in testing we achieved an accuracy score of 98.36%.
SAMPLE PREDICTIONS:
FOR TARGET:

FOR FAILURE TYPE:


COMPARISON:
● In the base paper,they have used an dataset consists of 10,000 data points stored as rows with 6 features in
columns,but we have included two more features in column which includes Target and Failure type, to
display a type of failure in an output.

● In the base paper, they have used an decision tree model to get the result,but we have used Random forest
model to get the results,random forest algorithm avoids and prevents overfitting by using multiple trees.
For large data,it produces highly accurate results.

● In the base paper,they have used Machine learning method and used the Explainable Artificial Intelligence to
check their ML model,but we have stopped till the Machine Learning model as our ML model has an high
accuracy score.XAI techniques in predictive maintenance applications can lead to increased complexity and
computational overhead, which can be challenging to manage and may result in slower decision-making.

● In the base paper, they have used 2 different temperature Air Temperature and Process Temperature but for
this project we have modified these and used as Temperature Difference.
CONCLUSION

❖ Although implementing predictive maintenance can be a challenging task, the correct tools
and technologies can make it simpler.

❖ This concept goes all the way to incorporation into an already-existing enterprise
application ,the scope is enormous..

❖ It is likely that we were using a research library to discover the best performing method.

❖ This approach won't pose a danger to data security when we consider deploying it in a
business context like spf.
❖ Since all of the data are being housed in those industries databases, we don't need to do
anything other than apply our model to the sector.

❖ We have to monitor the performance of the model over time and raise alarms if
accuracy drops below a minimum level.

❖ Tracking can be done using live data samples on a recreated model in a different
environment or in real time.

❖ A raised alarm might be a sign that the model needs to be updated or modified because
the structure it learnt from the data has changed
REFERENCES
[1] D. Gunning, "Explainable artificial intelligence (xai)." Defense Advanced Research Projects Agency (DARPA),
2017

[2] C. Molnar, “Interpretable machine Learning”,Interpretable Machine Learning - Christoph Molnar.

[3] G. McLachlan, “Discriminant analysis and statistical pattern recognition”. Wiley Series in Probability and
Mathematical Statistics: Applied Probability and Statistics. John Wiley & Sons, Inc., New York, 2004.

[4] S. Matzka, “Using Process Quality Prediction to increase Resource Efficiency in Manufacturing Processes”, First
International Conference on Artificial Intelligence for Industries (AI4I 2018), 2018

[5]S.Matzka, “AI4I 2020 Predictive Maintenance Dataset”, submitted to UCI Machine Learning Repository, 2020.

[6] Mathworks, “Estimates of predictor importance for classification ensemble of decision trees”
https://de.mathworks.com/help/stats/ compactclassificationensemble.predictorimportance.html, 2020.

[7] Ribeiro, Marco Tulio, Sameer Singh, and Carlos Guestrin. "Why should I trust you?: Explaining the predictions of
any classifier." Proceedings of the 22nd ACM SIGKDD international conference on knowledge discovery and data
mining. ACM, 2016
THANK YOU!!

You might also like