KR-20260062370-A - Payload Weight Predicter System of Excavator's Bucket based on Deep Learning

KR20260062370AKR 20260062370 AKR20260062370 AKR 20260062370AKR-20260062370-A

Abstract

This document is about a deep learning-based system that predicts the bucket load capacity of an excavator with mitigated operator bias. The proposed excavator bucket load capacity prediction system may include one or more sensors placed on the excavator; and a processor comprising an artificial neural network model configured to receive measurement values from the sensors and predict the load capacity. In this case, the artificial neural network model comprises: a feature extractor configured to extract features based on input values received from the sensors; a predictor configured to predict the load capacity based on the output values of the feature extractor; and one or more classifiers configured to implement adversarial training by outputting classification values for a domain and weight range that identify the operator based on the output values of the feature extractor.

Inventors

임정호
박경원

Assignees

에이치디한국조선해양 주식회사
에이치디현대인프라코어 주식회사
에이치디건설기계 주식회사

Dates

Publication Date: 20260507
Application Date: 20241029

Claims (10)

In a deep learning-based excavator bucket load capacity prediction system, One or more sensors disposed on the above excavator; and A processor comprising an artificial neural network model configured to receive the measurement value of the above sensor and predict the load weight, wherein The artificial neural network model mentioned above is, A feature extractor configured to extract features based on input values received from the above sensor; A predictor configured to predict the load weight based on the output value of the above feature extractor; and An excavator bucket load weight prediction system comprising one or more classifiers configured to implement adversarial training by outputting classification values for domains and weight ranges for identifying a driver based on the output values of the above feature extractors.
In Article 1, The artificial neural network model mentioned above is, The first error generated in the above predictor is reflected as is in the above input value, and An excavator bucket load weight prediction system that implements the adversarial learning by multiplying the second error occurring in one or more classifiers by a negative value and reflecting it in the input value.
In Article 1, The above one or more classifiers are, A domain classifier that classifies a domain identifying the above driver; and Excavator bucket loading weight prediction system including a weight classifier that classifies classification values for the above weight ranges.
In Paragraph 3, An excavator bucket loading weight prediction system that is learned by applying a first parameter having a negative value to the error value of the domain classifier and a second parameter having a negative value to the error value of the weight classifier to the input value.
In Article 4, An excavator bucket loading weight prediction system in which the first parameter and the second parameter are learned and updated during the deep learning process of the artificial neural network.
In Article 1, The measurement value of the above sensor is input into the above artificial neural network model after undergoing a preprocessing process, and The above preprocessing process is, Filtering is performed on the measurement value of the above sensor through a low-frequency filter; An excavator bucket loading weight prediction system comprising applying a sliding window to data filtered by the above low-frequency filter to generate a single sample of data for a predetermined period.
In Article 1, The above feature extractor and the above predictor include one or more of a CNN (Convolutional Neural Network) layer and an MLP (Multilayer Perceptron) layer, and An excavator bucket load capacity prediction system in which one or more of the above CNN layer and the above MLP layer are composed of a Bayesian Neural Network.
In Article 1, The above artificial neural network model is an excavator bucket load capacity prediction system composed of a CNN (Convolutional Neural Network) - LSTN (Long Short Term Memory) model.
In Article 1, The output data of the above predictor is, Excavator bucket load capacity prediction system outputting through a post-processing step applying uncertainty filtering by a Bayesian neural network.
In Article 1, An excavator bucket load capacity prediction system comprising the above artificial neural network model as a transformer-based model.

Description

Deep Learning-based Excavator Bucket Payload Prediction System The following description is about a deep learning-based excavator bucket load capacity prediction system, specifically a deep learning-based system that predicts the excavator bucket load capacity by mitigating the operator bias of the excavator. Regarding the prediction of excavator bucket load capacity, there has previously been a dynamics-based bucket load capacity prediction system as described in Published Patent Application No. 10-2005-0009102 (January 24, 2005). This document discloses a technology for measuring bucket weight by raising the boom in a specific area and considering changes in cylinder pressure, the temperature of the hydraulic fluid, etc. However, dynamic-based measurement systems have a problem where errors increase when moving outside a specific range. Additionally, dynamic-based measurement systems have the disadvantage of being difficult to measure in real-time when work is in progress continuously. To overcome these drawbacks, Patent No. 10-2561948 (July 27, 2023) proposed a system for predicting the weight of soil in an excavator bucket using a fine-tuning module of a transfer model utilizing a Genetic Algorithm (GA). In this document, a deep learning-based weight prediction system discloses a method of receiving sensor data attached to an excavator as a CAN signal and predicting the load weight using an artificial neural network model. However, these conventional deep learning-based measurement systems have a problem in that the error increases when predicting other drivers due to driver-specific bias. FIG. 1 is a diagram illustrating the background of a deep learning-based excavator bucket loading weight prediction system proposed in one embodiment of the present invention. FIG. 2 is a diagram illustrating the configuration of a deep learning-based excavator bucket loading weight prediction system according to an embodiment of the present invention. Figure 3 is a diagram specifically illustrating the learning process of an artificial neural network in the deep learning-based excavator bucket load weight prediction system of Figure 2. FIGS. 4 and 5 are drawings for explaining the concept of weight range bias and the results according to the embodiment of FIG. 3. FIG. 6 is a diagram illustrating the overall process of a deep learning-based excavator bucket loading weight prediction system according to an embodiment of the present invention. Figures 7 and 8 are drawings for explaining the preprocessing process of Figure 6 in detail. FIGS. 9 and FIGS. 10 are drawings for explaining the concept of a post-processing process using uncertainty filtering according to an embodiment of the present invention. Hereinafter, embodiments of the present invention are described in detail with reference to the attached drawings so that those skilled in the art can easily implement the invention. However, the present invention may be embodied in various different forms and is not limited to the embodiments described herein. Furthermore, in order to clearly explain the present invention in the drawings, parts unrelated to the explanation have been omitted, and similar parts throughout the specification are denoted by similar reference numerals. Throughout the specification, when a part is described as "including" a certain component, this means that, unless specifically stated otherwise, it does not exclude other components but may include additional components. FIG. 1 is a diagram illustrating the background of a deep learning-based excavator bucket loading weight prediction system proposed in one embodiment of the present invention. An excavator load weight prediction system according to one embodiment of the present invention, as illustrated in FIG. 1, advances from the conventional concept of simply performing weight measurement and pursues unmanned operation at construction sites. In the past, as described above, a dynamic model-based measurement system was used, requiring an IMU sensor for measurement, and a specific operating range was required for precise measurement. Due to the characteristics of construction machinery, it is desirable to configure it so that load weight can be predicted even during movement or rotation. To this end, one embodiment of the present invention proposes a machine learning technology that expands the measurement area and can be applied to existing models, thereby seeking to streamline work progress monitoring and equipment durability management. FIG. 2 is a diagram illustrating the configuration of a deep learning-based excavator bucket loading weight prediction system according to an embodiment of the present invention. An excavator bucket load weight prediction system according to one embodiment of the present invention may include one or more sensors (210) placed on an excavator (equipment) as shown in FIG. 2, and a processor (220) including an artificial neural network model (221) configured to