# Artificial Neural Network-Based Conveying Object Measurement Automation System Using Distance Sensor

## TL;DR

A low-cost sensor and machine learning can accurately measure object dimensions on a conveyor belt, offering a cost-effective alternative to expensive 3D scanners.

## Contribution

A systematic framework using an entry-level sensor and artificial neural network for reliable geometry measurement in logistics.

## Key findings

- The proposed method provides reliable object length and width measurements under varying transfer conditions.
- Entry-level sensors with feature engineering and machine learning can replace expensive 3D scanners for logistics inspection.
- The model handles measurement disturbances caused by changes in object location and logistics environment.

## Abstract

What is the main finding?
An entry-level sensor combined with moving-window features and an ANN can achieve accurate and stable geometry measurements on a conveyor, despite limitations in sensor resolution and beam characteristics.

An entry-level sensor combined with moving-window features and an ANN can achieve accurate and stable geometry measurements on a conveyor, despite limitations in sensor resolution and beam characteristics.

What are the implications of the main findings?
The results indicate that entry-level sensors, when paired with appropriate feature engineering and machine learning, can serve as a practical alternative to expensive 3D scanners for inline logistics inspection and box-dimension measurement.The identified limitations highlight the need for future models that explicitly account for variations in the conveyor speed, sampling rate, and distance between the sensor and object.

The results indicate that entry-level sensors, when paired with appropriate feature engineering and machine learning, can serve as a practical alternative to expensive 3D scanners for inline logistics inspection and box-dimension measurement.

The identified limitations highlight the need for future models that explicitly account for variations in the conveyor speed, sampling rate, and distance between the sensor and object.

Measuring technology is used in various ways in the logistics industry for defect inspection and loading optimization. Recently, in the context of the fourth industrial revolution, research has focused on measurement automation combining AI, IoT technologies, and measuring equipment. The 3D scanner used for field logistics measurements offers high performance and can handle large volumes quickly; however, its high unit price limits adoption across all lines. Entry-level sensors are challenging to use due to measurement reliability issues: their performance varies with changes in object location, shape, and logistics environment. To bridge this gap, this study proposes a systematic framework for geometry measurement that enables reliable length and width estimation using only a single entry-level distance sensor. We design and build a conveyor-belt-based data acquisition setup that emulates realistic logistics transfer scenarios and systematically varies transfer conditions to capture representative measurement disturbances. Based on the collected data, we perform robust feature extraction tailored to noisy, condition-dependent signals and train an artificial neural network to map sensor observations to geometric dimensions. We then verified the model’s performance in measuring object length and width using test data. The experimental results show that the proposed method provides reliable measurement results even under varying transfer conditions.