Project Progress Report on

COST REDUCTION AND PRODUCTIVITY

IMPROVEMENT THROUGH PREDICTIVE
MAINTENANCE IN MANUFACTURING

Submitted in partial fulfilment of the requirement for the award of the

degree of

BACHELOR OF TECHNOLOGY
IN
COMPUTER SCIENCE & ENGINEERING (B.TECH CSE)
Submitted by:

SARANG CHAUHAN University Roll No.-2019074

ABHISHEK CHAUHAN University Roll No.-2018624
VANSHAJ YADAV University Roll No.-2019290
NAVYA TRIPATHI University Roll No.-2018958

Under the Guidance of

Dr. Jyoti Agarwal
Associate Professor

Project Team ID: MP24CSE040

Project Progress Report No: 1

Department of Computer Science and Engineering

Graphic Era (Deemed to be University)
Dehradun, Uttarakhand
 2024-25

CANDIDATE’S DECLARATION
I/We hereby certify that the work which is being presented in the project progress report
entitled “Cost Reduction and Productivity Improvement through Predictive
Maintenance in Manufacturing” in partial fulfillment of the requirements for the award of
the Degree of Bachelor of Technology in Computer Science and Engineering ((B.Tech CSE)
in the Department of Computer Science and Engineering of the Graphic Era (Deemed to be
University), Dehradun shall be carried out by the undersigned under the supervision of Dr.
Jyoti Agarwal, Associate Professor, Department of Computer Science and Engineering,
Graphic Era (Deemed to be University), Dehradun.

SARANG CHAUHAN 2019074 signature

ABHISHEK CHAUHAN 2018624 signature
VANSHAJ YADAV 2019290 signature
NAVYA TRIPATHI 2018958 signature

The above mentioned students shall be working under the supervision of the undersigned on
the “Cost Reduction and Productivity improvement through Predictive Maintenance in
Manufacturing”

Supervisor Head of the Department

Examination

Name of the Examiners: Signature with Date

1.
2.

Table of Contents

Chapter No. Description Page No.

Chapter 1 Introduction and Problem Statement 1-3
Chapter 2 Objectives 4
Chapter 3 Project Work Carried Out 5-9
Chapter 4 Future Work Plan 10
Chapter 5 Weekly Task 11
References 12-13
Chapter 1

Introduction and Problem Statement

1.1 Introduction

The manufacturing sector is highly reliant on the smooth operation of machinery and
equipment to ensure consistent productivity. However, unplanned machine failures
can lead to significant downtimes, resulting in production losses, increased repair
costs, and operational inefficiencies. Many industries still rely on traditional
maintenance strategies, which are often based on guesswork and scheduled
maintenance intervals. These strategies include reactive maintenance, where repairs
are only performed after a machine breaks down, and preventive maintenance, where
servicing is done at regular intervals, regardless of the machine’s actual condition.
Relying heavily on guesswork can expose companies to unexpected equipment
failures and costly repairs. For mission-critical machinery, reducing uncertainties in
maintenance planning and budgeting is crucial. Guesswork in maintenance can lead to
increased financial and operational risks, leaving companies unprepared to deal with
unforeseen issues. This is where being proactive makes a significant difference, and
Predictive Maintenance (PdM) offers a solution.
Predictive Maintenance is a form of Condition-Based Monitoring (CBM) that
leverages machine-level sensors, Internet of Things (IoT) technology, and advanced
data analytics to continuously monitor machinery health. PdM systems can detect
early signs of machine failure, diagnose mechanical issues, and predict when
equipment is likely to fail in the future. By analyzing real-time sensor data such as
temperature, vibration, and pressure, PdM helps anticipate breakdowns before they
happen, ensuring that maintenance is performed only when necessary.
As estimated by John et al. in [1], PdM can significantly reduce maintenance
costs by shifting from reactive or preventive strategies to predictive strategies. Instead
of waiting for failures or performing unnecessary maintenance, predictive models
analyze historical and real time data to identify patterns indicating imminent
equipment failure. This proactive approach not only minimizes downtime but also
optimizes maintenance schedules, extending the lifespan of machines and improving
overall equipment effectiveness (OEE).

1
 The increasing integration of machine learning and IoT in manufacturing has
enhanced the capabilities of PdM systems, allowing for more accurate failure
predictions and enabling real-time monitoring of machinery health. Machine learning
models can analyze vast amounts of data, uncover hidden patterns, and provide
maintenance teams with actionable insights. The detailed review of related techniques
has been given in [2, 3], where machine learning algorithms like decision trees,
random forests, gradient boosting, and neural networks have been shown to improve
prediction accuracy
In summary, predictive maintenance is a key innovation for industries seeking to
improve their maintenance practices, reduce operational risks, and enhance machine
uptime through proactive monitoring and data-driven decision-making.

1.2 Problem Statement

The problem statement for the present work can be stated as follows:
In a manufacturing environment, the unplanned breakdown of machinery and
equipment can result in costly repairs, lost production time, and operational
inefficiencies. Traditional maintenance strategies, such as reactive and preventive
maintenance, often fall short in predicting machine failures, leading to either over-
maintenance or unexpected breakdowns.
The goal of the present work is to develop a Predictive Maintenance (PdM)
system that utilizes machine learning algorithms to predict equipment failures or the
need for maintenance before they occur. By analyzing real-time sensor data from
machines and historical maintenance records, this system will be able to predict when
failures are likely to happen and allow for maintenance to be performed proactively,
reducing unplanned downtime.
Predictive Maintenance will help overcome the limitations of traditional
maintenance approaches by continuously monitoring the health of machines through
sensor data such as temperature, pressure, and vibration. Machine learning models
will be employed to predict potential failures and optimize maintenance schedules.
This system will be able to detect early signs of failure, provide warnings, and
recommend the best time for maintenance, ensuring that the machines continue to
operate without unnecessary interruptions. By implementing this system, the project
aims to:

2
 Minimize unplanned downtimes
 Reduce maintenance costs by optimizing repair schedules.
 Improve overall equipment effectiveness (OEE) by providing data-driven insights
into the condition of the machines.
 The proposed system will be capable of real-time data ingestion, anomaly detection,
and model-driven failure prediction, ultimately assisting maintenance teams in
making informed decisions on when and how to perform maintenance. The proactive
nature of predictive maintenance will not only reduce financial risks but also enhance
productivity and efficiency in manufacturing operations.

3
Chapter 2

Objectives
The goals of the suggested work on Predictive Maintenance in Manufacturing are mentioned
below:

1-To investigate the ideas of predictive maintenance in manufacturing as to its importance,

trends, and scope of application of data analytics methods for enhancing operational
efficiency.

2-To map the critical issues and knowledge voids in the developed predictive maintenance
models in the context of manufacturing industry and justify those gaps using appropriate
concepts.

3-To create a predictive maintenance model utilizing machine learning technology, which
aims to predict the breakdown of equipment and enable timely maintenance interventions
relying on real-time insights.

4- To encourage a change from the conventional "fix when broken" or "fix before break"
maintenance methods to a more effective approach that puts analytics at the core in order to
reduce expenses and improve equipment performance with the least amount of OEE.

4
Chapter 3

Project Work Carried Out

1. Data Requirements
Input Data
1. Dataset: A dataset containing sensor readings, operational metrics, and machine
failure records.
2. Features: Features include temperature, pressure, vibration levels, rotational speed,
and torque values.
3. Target: A binary target column indicating whether machine failure has occurred (1)
or not (0).
Output Data
1. Machine Failure Prediction: Binary classification results categorizing whether a
machine is likely to fail or not.
2. Model Performance Metrics: Metrics for evaluation, including accuracy, AUC,
precision, recall, F1 score, and F2 score.

2. Algorithm Approach
Phase 1: Data Processing
1. Data Loading & Cleaning: Load sensor data into a DataFrame and handle missing
or inconsistent values.
2. Preprocessing: Standardize numerical features, encode categorical variables (if any),
and split data into training, validation, and testing sets using an 80/10/10 ratio.
3. EDA (Exploratory Data Analysis): Visualize feature distributions, relationships, and
detect outliers using histograms, boxplots, and correlation heatmaps.
Phase 2: Model Training
1. Logistic Regression: Logistic Regression predicts the probability of binary outcomes
using the sigmoid function and optimizes parameters using gradient descent.

Table 3.1 Pseudo code of the Logistic Regression algorithm

5
Input:
D - the dataset, α - learning rate, N - number of iterations

begin
1. Initialize weights W randomly;
2. Initialize bias b = 0;
3. for i = 1 to N do
Compute predictions Y_pred = sigmoid (W·X + b);
Compute gradients:
dW = (1/m) * Σ (Xi * (Y_pred - Yi));
db = (1/m) * Σ (Y_pred - Yi);
Update parameters:
W = W - α * dW;
b = b - α * db;
4. end for
5. Output W and b
End

2. K-Nearest Neighbors (KNN): K-Nearest Neighbors assigns a class to a data point by

analyzing the classes of its k nearest neighbors in the dataset based on distance
measures like Euclidean distance.
Table 3.2 Pseudo code of the KNN algorithm

Input:
D - the dataset, k - number of neighbors, X_new - input data point to classify
begin
1.For each data point in D:
Calculate distance d (X_new, Xi) for all points in D;
2.Sort distances in ascending order;
3.Select k-nearest neighbors;
4.Assign the majority class among the k-nearest neighbors to X_new;
5.Output the assigned class for X_new
End

6
3. Support Vector Classifier (SVC): Support Vector Classifier finds a hyperplane in an
N-dimensional space that separates the data points into classes while maximizing the
margin between them.

Table 3.3 Pseudo code of the SVC algorithm

Input:
D - the dataset, C - regularization parameter, N - number of iterations

begin
1. Initialize weights W and bias b randomly;
2. for i = 1 to N do
3. for each (Xi, Yi) in D do
4. if Yi * (W·Xi + b) < 1 then
5. Update W = W - α * (2 * λ * W - Yi * Xi);
6. Update b = b + α * Yi;
7. else
8. Update W = W - α * (2 * λ * W);
9. end if
10. end for
11. end for
12. Output W and b
End

4. Random Forest Classifier (RFC): Random Forest constructs an ensemble of

decision trees and combines their predictions using majority voting for classification.

Table 3.4 Pseudo code of the RFC algorithm

Input:
D - the dataset, T - number of trees, m - number of features for splits

begin
1. Initialize RandomForest as an empty ensemble;
2. for t = 1 to T do

7
 3. Generate a random sample D_t with replacement from D;
4. Train a decision tree T_t on D_t:
5. At each node:
6. Select m random features;
7. Find the best feature and split point using Gini impurity;
8. Stop growing when a stopping condition is met;
9. Add T_t to RandomForest;
10. end for
11. Output the RandomForest ensemble
End

5. Extreme Gradient Boosting (XGB): XGBoost uses gradient-boosted decision trees to

minimize errors by focusing on residuals from previous trees in an iterative fashion.

Table 3.5 Pseudo code of the XGB algorithm

Input:
D - the dataset, N - number of iterations, η - learning rate

begin
1. Initialize predictions F0(X) = mean (target values in D);
2. for i = 1 to N do
3. Compute residuals r = Yi - Fi-1(Xi);
4. Fit a regression tree Ti to predict residuals r;
5. Update predictions:
6. Fi(X) = Fi-1(X) + η * Ti(X);
7. end for
8. Output the final model F(X)
End

Phase 3: Machine Failure Prediction

1. Performance Metrics Evaluation: Evaluate all models using metrics like accuracy,
AUC, precision, recall, F1 score, and F2 score to identify the best-performing model.
2. Confusion Matrices Analysis: Generate confusion matrices for each model to
analyze classification performance in detail.

8
 True Positives (TP): Correctly predicted failures.
True Negatives (TN): Correctly predicted non-failures.
False Positives (FP): Non-failures incorrectly predicted as failures.
False Negatives (FN): Failures missed by the model.
3. Model Comparison: Compare the performance of all trained models (Logistic
Regression, KNN, SVC, RFC, XGB) to determine the most suitable one for predicting
machine failures.

Table 3.6 Validation Scores

Algorithm Accurac Area Under F1 Score F2 Score

y (ACC) Curve (AUC)
K Nearest Neighbor 0.959 0.954 0.902 0.928
Support Vector Classifier 0.966 0.987 0.916 0.931
Random Forest Classifier 0.977 0.997 0.943 0.954
Extreme Gradient Boosting 0.987 0.999 0.969 0.970
Logistic Regression 0.905 0.883 0.673 0.629

Table 3.7 Test Scores

Algorithm Accuracy Area Under F1 Score F2 Score

(ACC) Curve (AUC)

Logistic Regression 0.881 0.913 0.655 0.598

K Nearest Neighbor 0.966 0.954 0.916 0.927
Support Vector Classifier 0.973 0.992 0.934 0.941
Random Forest Classifier 0.972 0.997 0.931 0.945
Extreme GradientBoosting 0.983 0.998 0.958 0.956

9
Chapter 4
Future Work Plan

The future work plan of our project are as follows:

Sl. No. Work Description Duration in Days

1. Experiment with ensemble learning methods 14

combining Logistic Regression, KNN, and
SVM for improved accuracy.
2. Evaluate the performance of ensemble models 14
by combining Random Forest and XGBoost to
improve model robustness.
3. Tune hyperparameters of models (Logistic 14
Regression, SVM, Random Forest, XGBoost)
to optimize performance metrics.
4. Experiment with CNN and LSTM for 14
predictive maintenance in manufacturing, using
sensor data and equipment history to predict
failure and remaining useful life.
5. Perform model validation using cross- 10
validation and fine-tune the final model
selection based on precision, recall, and F1-
score.
6. Select major features based on importance 10
scores and correlation to improve model
interpretability and performance.

10
Chapter 5

Weekly Task

The report of project work allocated by the supervisor is as follows:

Work
Compl
Wee Date: Remar Guide
Work Allocated eted
k No. From-To ks Signature
(Yes/
No)
1 20-10-2024 to Load the dataset and clean Yes
26-10-2024 unnecessary columns, handle
missing data.
2 27-10-2024 to Preprocess the data (scaling, Yes
02-11-2024 encoding,categorical variables,
split into training, validation,
and test sets).
3 03-11-2024 to Implement and train Logistic Yes
09-11-2024 Regression and KNN models
4 10-11-2024 to Implement and train Support Yes
16-11-2024 Vector Classifier (SVC) and
Random Forest models.
5 17-11-2024 to Implement and train Extreme Yes
23-11-2024 Gradient Boosting (XGB).
6. 24-11-2024 to Evaluate models using Yes
30-11-2024 performance metrics
7. 01-12-2024 to Generate confusion matrices Yes
03-12-2024 and analyze misclassifications
(TP, TN, FP, FN).
8. 04-12-2024 Compare the performance of Partial
all models and finalize the best
model for machine failure
prediction.

References
11
[1] T. P. Carvalho, F. A. Soares, R. Vita, R. D. P. Francisco, J. P. Basto, and S. G. Alcalá, "A
systematic literature review of machine learning methods applied to predictive
maintenance," Computers & Industrial Engineering, vol. 137, no.1, pp. 106024, 2019.

[2] P. Aivaliotis, K. Georgoulias, and G. Chryssolouris, "The use of Digital Twin for
predictive
maintenance in manufacturing," International Journal of Computer Integrated
Manufacturing, vol. 32, no. 11, pp. 1067-1080, 2019.

[3] W. Yu, T. Dillon, F. Mostafa, W. Rahayu, and Y. Liu, "A global manufacturing big data
ecosystem for fault detection in predictive maintenance," IEEE Transactions on Industrial
Informatics, vol. 16, no. 1, pp. 183-192, 2019.

[4] S. T. March and G. D. Scudder, "Predictive maintenance: strategic use of IT in

manufacturing organizations," Information Systems Frontiers, vol. 21,no. 1, pp. 327-341,
2019.

[5] Z. M. Çınar, A. Abdussalam Nuhu, Q. Zeeshan, O. Korhan, M. Asmael, and B. Safaei,

"Machine learning in predictive maintenance towards sustainable smart manufacturing in
industry 4.0," Sustainability, vol. 12, no. 19, pp. 8211, 2020.

[6] T. Zonta, C. A. Da Costa, R. da Rosa Righi, M. J. de Lima, E. S. da Trindade, and G. P.

Li,
"Predictive maintenance in the Industry 4.0: A systematic literature review," Computers
&
Industrial Engineering, vol. 150, no.1, pp.106-189, 2020.

[7] J. Lee, J. Ni, J. Singh, B. Jiang, M. Azamfar, and J. Feng, "Intelligent maintenance
systems and predictive manufacturing," Journal of Manufacturing Science and
Engineering, vol. 142, no. 11, pp. 110805, 2020.

[8] C. Krupitzer, T. Wagenhals, M. Züfle, V. Lesch, D. Schäfer, A. Mozaffarin, J. Edinger, C.

Becker, and S. Kounev, "A survey on predictive maintenance for industry 4.0," preprint at
Xiv, vol.202, no.1, pp. 182-24, 2020.

12
[9] B. Hrnjica and S. Softic, "Explainable AI in manufacturing: a predictive maintenance case
study,"IFIP International Conference on Advances in Production Management Systems,
Cham: Springer International Publishing, vol.1, no.1, pp. 66-73. 2020.

[10] M. Pech, J. Vrchota, and J. Bednář, "Predictive maintenance and intelligent sensors in
smart factory," Sensors, vol. 21, no. 4, pp. 1470, 2021.

[11] S. Ayvaz and K. Alpay, "Predictive maintenance system for production lines in
manufacturing: A machine learning approach using IoT data in real-time," Expert
Systems with Applications, vol. 173, no.1, pp. 114598, 2021.

[12] G. M. Sang, L. Xu, and P. de Vrieze, "A predictive maintenance model for flexible
manufacturing in the context of industry 4.0," Frontiers in Big Data, vol. 4, pp. 663466,
2021.

[13] C. Liu, D. Tang, H. Zhu, and Q. Nie, "A novel predictive maintenance method based on
deep adversarial learning in the intelligent manufacturing system," IEEE Access, vol. 9,
pp. 49557-49575, 2021.

[14] X. Han, Z. Wang, M. Xie, Y. He, Y. Li, and W. Wang, "Remaining useful life prediction
and predictive maintenance strategies for multi-state manufacturing systems considering
functional dependence," Reliability Engineering & System Safety, vol. 210, pp. 107560,
2021.

[15] H. Choi, D. Kim, J. Kim, J. Kim, and P. Kang, "Explainable anomaly detection
framework for predictive maintenance in manufacturing systems," Applied Soft
Computing, vol. 125, pp. 109147, 2023.

13
14