Feature Extraction for Digging Operation of Excavator

Improvement of the work efficiency is demanded by aging and reducing of the working population in the construction field, so that some automation technologies are applied to construction equipment, such as bulldozers and excavators. However, not only the automation technologies but also expert skills are necessary to improve the work efficiency. In this paper, the human skill evaluation is proposed by the human skill evaluation of the data-driven PID controller. The proposed method is applied to the excavator digging operation. As the result, the difference between the novice operation and the skilled operation is extracted. Moreover, the numerical difference is clarified based on the result.


Introduction
Japanese Ministry of Land, Infrastructure, Transport and Tourism have started the policy of the integration of construction and ICT (Information and Communication Technology) for public works such as land survey, engineering, construction and maintenance, that is called "i-construction". 1 Aims of this policy are the improvement of the work efficiency and the increasing of productivity.The technologies of IoT (Internet of Things), big data and AI (Artificial Intelligence) will be applied to some construction equipments based on the policy.However, human operations for construction equipments are still required continuously in the field because skilled workers optimize their operations for high productivity in each complicated conditions.Unfortunately, those skilled operators who have the unique technique are reduced in these days because of the aging and reducing of the working population.
As the previous studies, skill based controllers have been proposed. 2,3These studies considered that the human skill described as a controller with the tuner of PID gains, called the skill-based PID controller.The operation skill have been replicated by the skill-based PID controller experimentally.Moreover, the skill evaluation by using the skill-based PID controller have been performed. 4,5These research focus on the change of PID gains to extract the feature of the operation skill for the excavator.However, the evaluated operation is applied to the simple lever operation such as the swing operation.In the actual work field, the operation for the excavator is more complicated.
In this study, the data-driven skill-based PID controller is adopted to the actual work operation of the excavator, i.e. the continuous digging operation.The skill feature for the excavator of the continuous digging operation is extracted to compare the novice operation with the P -425 professional operation based on the control engineering approach.Moreover, the numerical difference of feature between is verified.Algorism of Skill Feature Extraction

Scheme of skill feature extraction
The scheme of feature extraction for the human skill is shown in Fig. 1.Based on the engineering approach, the operational objective and the operator are considered as the system and the PID controller with the parameter tuner that is consist of the data driven method.* ( ) and * ( ) are denoted as the system output and the system input by the operation.

Data-driven PID controller
The following control law is considered for the skillbased controller: where the PID gains, ( ), ( ) and ( ) are tuned by the database.The control error signal ( ) is defined as ( ): = ( ) − ( ) and the difference operator ∆ is defined as ∆≔ 1 − .The operational objective is considered as the nonlinear system is shown as follows: Here the ℎ(•) is the nonlinear function using past data and the information vector ( − 1) is defined as: where and orders of output and input, respectively.The query ( ) is defined as: The dataset of database, ( ), is constructed as follows: where the index and the total number of datasets is shown as i and N, respectively.PID gains that is requested by the query are calculated by the following procedure.Distances (•) between the query ( ) and ( )in each datasets are calculated by: where the index of an element in a dataset or query is shown as l.
is a set of the element for the index.Neighbors of data are selected according to their distance from query.The number of neighbor dataset, that is the user specified parameter, is donated as k.The sum of selected PID gains ( ) multiplied by a weight is desired PID gains ( ) as follows: The following steepest descent method (10) is performed for the learning of PID gains in the database.
where the error criterion ( ) ,the updating gradient ( ) and the partial differential are expressed as follows: Coefficients of a, b and c are the positive

Experimental equipment
The proposed scheme is applied to the operation of an excavator.The evaluated operation is a continuous digging motion and the excavator, SK200-9 made in Kobelco Construction Machinery, is used for the experiment.The continuous digging motion can be classified into four parts in Fig. 3. First part is called digging that is the operation to scoop the earth and sand with the bucket ①.Then, the load is lifted to the certain height that assumed a carrier of the truck ②.Furthermore, the load is dumped on the ground that assumed the truck bed ③.Finally, teeth of the bucket is repositioned to the starting point of the digging ④.

Skill evaluation procedure
The most important operation for actual works is the positioning at the stop motion because the skill is required by the operation.Therefore, the target for this evaluation is the positioning by the stop operation only.The acceleration should be ignored in this paper.The evaluation procedure is shown as follows.The proposed method is applied to the specified operation into the continuous digging operation.In this study, the swing operation is selected as the specified operation because the skill is necessary to stop at the proper position and the reference value ( ) of operation is clearer than another operation.The digging operation is performed to record * ( ) and ( ).Here, the pilot pressure and the angle of the swing are set as the input value * ( ) and the output value * ( ), respectively.Then, the skill feature is extracted by the comparison of the professional and novice operation.Moreover, the physical quantity that can be measured is evaluated based on the skill feature.

Application result of proposed method
The data of the swing motion in the continuous digging operation is extracted, normalized and resampled by the sampling time of 200ms according to the human quickness of a response.The proposed method is applied to the professional and novice data.The user specific parameter of the proposed method is shown as follows.The calculation results of PID parameter for the professional operator and the novice operator are shown in Fig. 4 and Fig. 5.As the result, the change of PID gain at the period of deceleration which needs the operation skill has the opposite properties to compare with them.This is the almost same result of the skill evaluation for the single operation.Refer to the reference 4 in the detail of the single operation.The result shows that the professional operator can be expressed as the proper nonlinear controller and the novice operator can be expressed as the ON/OFF controller.

Skill numerical evaluation
The difference between the ON/OFF controller and the proper nonlinear controller is considered based on the skill feature.Typically, ON/OFF controller has the larger input value compared with the proper controller, that is, the novice operator has the larger operational input.The change of input expresses one of the skill criterion for the excavator operation.The Table 2 shows the change of the input signal for the 10 cycle of the continuous digging operation.As the result, it is verified that the novice has the unnecessary input compared with the professional.

Conclusion
The human skill evaluation based on the data-driven PID controller has been performed to the excavator operation of the continuous digging motion.The difference between the novice and professional is verified numerically on the skill feature.Especially, the skill feature that focuses on the quantity of input signal based on the control engineering approach is extracted numerically.In future work, statistical and stochastic approach will be applied by the increasing of data.

1 . 3 .Fig. 1
scheme in Fig.2.PID controller, which has the change of PID gains can be expressed as the feature of the human skill.