Adaptive Payload Estimation: A Universal Payload Estimation System for Wheel Loaders Using Recurrent Neural Networks and Transfer Learning
Typ
Examensarbete för masterexamen
Master's Thesis
Master's Thesis
Program
Information and communication technology (MPICT), MSc
Complex adaptive systems (MPCAS), MSc
Complex adaptive systems (MPCAS), MSc
Publicerad
2024
Författare
Carlsson, Marcus
Hiljemark, Elin
Modellbyggare
Tidskriftstitel
ISSN
Volymtitel
Utgivare
Sammanfattning
A wheel loader is a versatile construction machine designed for various tasks related
to the loading and transport of materials. In many wheel loader applications, accurately knowing the mass of the load being lifted is crucial. For example, when
loading materials onto trucks, it is essential to ensure that the load of the truck does
not exceed legal weight limits. Therefore, various solutions have been developed to
help operators determine the payload mass. However, many existing systems use
mechanical equations to estimate the payload, consequently these models are heavily
dependent on the geometry of the machine it is implemented for. Using technologies
emerging from the field of deep learning, this thesis aimed to investigate the possibilities of a universal payload estimation model. By leveraging real operational data
from wheel loaders and formatting it into time sequences, a neural network based on
long short-term memory units was constructed. In addition, a classification network
was developed based on both convolutional neural network and autoencoder architectures to assess the reliability of data sequences. To generalize the model across
different wheel loader models, transfer learning was employed. The results showed
that transfer learning enabled the model to accurately estimate payload mass for
different wheel loader models. However, the classification network for sensor data
reliability was found to be redundant, as it reduced the performance of the payload estimation model. Furthermore, the result also showed a trade-off between
the amount of data used for fine-tuning the general model and the accuracy of the
model. To further investigate the adaptability and usage of the adaptive payload
estimation model, more operational scenarios needs to be tested as well as further
examination of the adaptability to different sizes of wheel loaders.
Beskrivning
Ämne/nyckelord
Transfer learning , LSTM , payload estimation , wheel loader , deep learning , load weighing , construction equipment