Kiln Process Fan Vibrations Prediction Based on Machine Learning Models: Application to the Raw Mill Fan ^†

Abstract

Kiln process fans play a crucial role in the cement manufacturing process. This article presents a study on the use of machine learning models to predict kiln process fan vibrations based on process fan running parameters. The study tested three different models, namely, k-nearest neighbors, linear regression, and random forest, and it found that all three could accurately predict fan vibrations. However, the random forest model performed the best due to its ability to handle non-linear relationships. The findings have significant implications for the maintenance and operation of kiln process fans, as predictive maintenance can reduce downtime and improve operational efficiency.

1. Introduction

Industrial kilns are widely used in various manufacturing processes, such as cement production, ceramics manufacturing, and chemical processing [1]. However, the operation of the kiln process fan can be affected by various factors, leading to excessive vibrations and potential breakdowns. Therefore, accurate prediction of fan vibrations is essential for improving the reliability and efficiency of kiln systems. In recent years, machine learning (ML) techniques have shown great potential in predicting industrial equipment vibrations [2]. This article explores the application of ML models in predicting kiln process fan vibrations and discusses their effectiveness in improving kiln system performance.

2. Kiln Process Fans

Kiln process fans are used in the cement and minerals processing industries to move hot gases and materials through a rotary kiln. In a cement factory, the six main process fans are [3]:

• Raw mill fan;
• Preheater fan;
• Kiln fan;
• Clinker cooler fan;
• Coal mill fan;
• Cement mill fan.

In summary, each of the six main process fans in a cement factory plays a critical role in the cement manufacturing process. In the rest of this paper, the study focuses on the raw mill fan as a BC process fan.

3. Process Fan Running Parameters

The running parameters of a process fan can vary depending on various factors such as the type of fan, operating conditions, and the specific application. For instance, in our case of the raw mill process showed in Figure 1, the running parameters of the raw mill fan may include:

• Outlet gas temperature: the temperature of the gas air flow leaving the kiln;
• Fan speed: the speed of the fan that delivers the combustion air to the kiln;
• Stage DE temperature: drive-end fan bearing temperature;
• Stage NDE temperature: non-drive-end fan bearing temperature;
• Fan motor intensity: the amount of electrical power consumed by the fan motor;
• Draft fan airflow: the volume of air or air flow gas moved by the fan that creates the draft.

These parameters cited in

Table 1. Raw mill fan running parameters.

Active Parameters	Unit
BC outlet gas temperature	°C
BC fan speed	RPM
Stage DE Temperature	°C
Stage NDE Temperature	°C
Fan motor intensity	AMP
Draft fan airflow	Nm3/h

are critical for the efficient and consistent operation of the kiln process and are typically monitored and controlled using advanced automation and control systems.

4. Process Fan Vibrations Acquisition

Acquiring process fan vibration data can be useful in detecting potential mechanical issues or faults that may lead to failure or decreased performance of the fan. Here are the steps to acquire process fan vibration data and store it in the database [4,5]:

• Identify the location where the vibration data needs to be collected;
• Install the vibration sensors;
• Connect the sensors to the data acquisition system;
• Choose a database management system (DBMS) to store the vibration data;
• Configure data acquisition system to collect vibration data at the desired frequency and amplitude ranges, as well as to store the data in the database;
• Store data in the database in real-time.

5. Case Study

In our case study, we will work with data of a raw mill fan.

5.1. Data Presentation

Our process fan data is presented in two categories; the first one is raw mill fan running parameters, as

Table 2. Raw mill fan running parameters data.

	Date (nbr)	Outlet Gaz Temp	Fan Speed	DE Bearing Temp	NDE Bearing Temp	Motor Intensity	Fan Air Flow
0	43,956.375000	73.12	916.34	33.16	44.91	1202.36	228,393.15
1	43,956.416667	74.69	916.34	34.11	48.83	1201.62	226,188.93
2	43,956.458333	74.64	916.34	35.35	51.85	1190.90	225,487.57
3	43,956.500000	74.85	916.26	36.36	53.62	1182.79	224,862.90
4	43,956.541667	75.01	916.13	37.60	54.57	1186.48	225,946.48

shows, and the second one is raw mill fan vibrations, as

Table 3. Raw mill fan vibrations data.

	mm/s	G	gE	Date (nbr)
0	0.513	0.0460	0.022	44,208.328368
1	0.432	0.0455	0.023	44,208.222396
2	8.596	1.2947	1.091	44,208.087373
3	28.783	4.9991	2.361	44,208.086366
4	24.697	5.1883	2.142	44,208.082905

shows.

5.2. Data Preprocessing

Data preprocessing refers to the process of cleaning, transforming, and preparing raw data to make it suitable for analysis. It is an essential step in data analysis as raw data often contains inconsistencies, errors, and missing values that can affect the accuracy of the analysis.

5.2.1. Data Cleaning

• Identify and handle missing values;
• Identify and handle outliers;
• Identify and handle inconsistent data.

5.2.2. Data Integration

We merge any relevant datasets that contain raw mill fan running parameters and raw mill fan vibrations, and

Table 4. Data Integrated.

	Date (nbr)	Outlet Gaz Temp	Fan Speed	DE Bearing Temp	NDE Bearing Temp	Motor Intensity	Fan Air Flow	Vibration mm/s	Vibration G	Vibration gE
0	43,956.375000	73.12	916.34	33.16	44.91	1202.36	228,393.15	0.513	0.0460	0.022
1	43,956.416667	74.69	916.34	34.11	48.83	1201.62	22,6188.93	0.432	0.0455	0.023
2	43,956.458333	74.64	916.34	35.35	51.85	1190.90	22,5487.57	8.596	1.2947	1.091
3	43,956.500000	74.85	916.26	36.36	53.62	1182.79	22,4862.90	28.783	4.9991	2.361
4	43,956.541667	75.01	916.13	37.60	54.57	1186.48	22,5946.48	24.697	5.1883	2.142

shows our data after integration.

5.3. Modeling and Evaluation

The three models that we used to predict kiln raw mill fan vibrations based on process parameters are [6,7]: linear regression, k-nearest neighbors, and random forest.

When selecting machine learning models to predict kiln process fan vibrations in our study, we have considered several commonly used criteria. These include the simplicity and interpretability of the models, their ability to handle non-linear relationships and high-dimensional data, their performance and accuracy, and the availability of computational resources.

To evaluate the performance of an ML model for the prediction of continuous values, there are several metrics that can be used. Using both R² and RMSE as evaluation metrics provides a more comprehensive picture of the model’s performance. R² measures how well the model fits the data overall, while RMSE measures the error between the predicted and actual values, providing insight into how well the model can predict future values.

• Root Mean Squared Error (RMSE)

$$RMSE=\sqrt{\frac{1}{n}{{\displaystyle \sum }}_{i=1}^n{\left(y_i-{\widehat{y}}_i\right)}^2}$$

(1)

• R-squared (R²)

$$R^2=1-\frac{{\sum }_{i=1}^n{\left(y_i-{\widehat{y}}_i\right)}^2}{{\sum }_{i=1}^n{\left(y_i-\overline{y}\right)}^2}$$

(2)

where:

n is the number of data points;

$y_i$ is the actual value for the i-th data point;

${\widehat{y}}_i$ is the predicted value for the i-th data point;

$\overline{y}$ is the mean of the actual values in the data.

5.4. Results

Table 5. Performances evaluation.

	R2	RMSE
Linear Regression	46.77%	1.62
kNN	64.88%	1.36
Random Forest	72.38%	1.21

shows the performance of three different machine learning models on a given dataset. The models used are k-nearest neighbors (kNN), random forest, and linear regression.

Random Forest has the best performance in terms of R² and RMSE. It has an R² value of 72.38%, which means that 72.38% of the variance in the target variable can be explained by the independent variables in the model. It also has the lowest RMSE value of 1.21, indicating that it has the lowest amount of error between the predicted and actual values.

Overall, these results suggest that Random Forest is the best model for this dataset, based on its higher R² and lower RMSE values.

6. Conclusions

In conclusion, the objective of this work was to predict the kiln process fan vibrations using machine learning models in order to improve the cement kiln overall performance. In fact, the random forest model is the best model found, with an R² value of 72.38% and an RMSE value of 1.21. These statistics indicate that the model can accurately explain 72.38% of the variance in the target variable and has the lowest amount of error between predicted and actual values. The implementation of our model in industrial settings will have significant implications for maintenance and operation by allowing operators to identify potential issues before they become critical and perform predictive maintenance, and it will help to improve operational efficiency and introduce cost savings for the cement plant.