Indonesian Waste Database: Smart Mechatronics System

Reuse


Introduction
In recent years, the trash problem has become a significant concern [1]- [3]. Even during the COVID-19 epidemic, waste output has increased dramatically, particularly medical waste [4]- [6]. In general, waste is classified as organic or inorganic. Organic waste is garbage made from natural elements that can be immediately reduced and degraded [7], [8]. Meanwhile, inorganic waste is defined as garbage-containing compounds that are difficult to degrade in the environment [9].
Waste may be decreased by employing the 3R concept (Reduce, Recycle, Reuse) [10]- [12]. Waste reduction can be addressed by minimizing items with the potential to become waste. Furthermore, waste reduction might use products already used for reuse. In reality, waste management may be recycled to create trash with new purposes and a higher value [2], [13]- [15].
Indonesia's waste management system is still far from ideal [16]- [18]. In Indonesia, pollution from trash and discarded limbs are at an all-time low. In Indonesia, waste is classified based on the primary substance. These wastes might be paper, cardboard, plastic, medical waste, organic trash, or metal.
Today, the role of technology in dealing with waste concerns is critical. Much research has been conducted in recent years to promote the use of technology in trash management. The Internet of Things [19]- [21], robotics [22]- [25], automation [26]- [28], and artificial intelligence [29]- [31] are some of the technologies that may be employed in waste management. The Internet of Things can play a role in the management of waste information in the environment as well as a centralized means of delivering data. At the same time, robots and automation may be employed as waste management actuators in the environment. The mechatronics method refers to the integration of robotics and automation.
Artificial Intelligence, the most recent form of waste management, provides advantages regarding system accuracy. In recent years, artificial intelligence has been used to identify garbage categories, detect rubbish objects, and anticipate waste. By applying computer-based [32], mobile [31], and web-based applications [33], artificial intelligence may also be employed as an integrated system. A database that can be utilized as a learning system is essential in artificial intelligence. This database system is one-of-a-kind and restricted to specified requirements. The location of the garbage, the kind of waste, and the characteristics of the waste are among these criteria.
As a result, the research's objective and originality are to categorize Indonesian garbage using a mix of artificial intelligence and mechatronics technologies. First, an Indonesian trash database is constructed based on the material characteristics and the kind of features utilized. In addition, DenseNet artificial intelligence learning is used to get the optimal accuracy value. Simultaneously, the mechatronics arm robot system is finally developed with the highest precision of artificial intelligence output model.

Method
This research is a product-based project. In order to accommodate each step of research, the ADDIE (Analyze, Design, Develop, Implement, Evaluate) technique is employed [34], [35].

Analyze
Environmental research and a search for waste characteristics in Indonesia were used in the analysis. The Indonesian trash database was created in numerous categories: paper, cardboard, plastic, medical waste, organic garbage, and metal. Table 1 shows the distribution class of the Indonesian waste database. The total number of photos in Indonesia's waste database is 1880. The training phase consumes 80% of the database. Furthermore, testing and validation consume 20% of the waste database.

Design
The design step includes product simulation and mechatronics planning. This step involves developing a mobile robot software prototype with a mechatronics system. The robot features four wheels and a robotic arm-based servo actuator with three degrees of freedom (DOF). The robotic arm in this project was inspired by Tung's [36] research that decreased the actual system's DOF. The 3D design of the mechatronics system, on the other hand, may be shown in Fig. 1 (a)-(b).

Development
This research combines two approaches, mechatronics systems and artificial intelligence. Mechatronics systems are used as drive actuators. Meanwhile, artificial intelligence is used as an input to the state of the environment using image detection. In general, the system flowchart can be seen in Fig. 2.

Mechatronics Systems
Mechatronics-based systems are system techniques that incorporate electrical, software, and mechanical actuators [37]- [39]. At the Development stage, Mechatronics robot manufacturing is carried out with an approach to adding hardware and electrical circuits. As a mechatronics system robotic arm actuator, servo hardware is employed. The camera is also installed on the robotic arm as

Artificial Intelligence
DenseNet was employed as artificial intelligence in this investigation. DenseNet is a deep learning system based on dense layers [40]. DenseNet has numerous algorithm variations dependent on depth and layer count. DenseNet has the benefit of being simple to implement and accurate. In addition to DenseNet, numerous related algorithms, including as MobileNet [41]- [43], Single Shot Detector (SSD) [44]- [46], Yolo [47]- [50], and ResNet [51]- [53], can be used in specific case studies.

Implementation
At this point, the software and hardware components are combined by installing the Jetson Nano as the primary CPU. Jetson Nano supports artificial intelligence data transmission with a robotic system based on mechatronics. The whole wiring diagram can be seen in Fig. 4. At the implementation stage, system detection testing is carried out using the camera contained in the robot. The results of camera detection tests with Jetson Nano can be seen in Fig. 5.

Evaluate
The last stage is evaluation as a system improvement. At this evaluation stage, several optimizations and accuracy comparisons are carried out. The evaluation stage was deducted to find the best output model for the mechatronics system. In the evaluation phase, the effectiveness of integrating artificial intelligence and mechatronic systems can be observed.

Results and Discussion
This study used a model of ingenuity made by the DenseNet algorithm, as shown in Fig. 6. All types of artificial intelligence are used as a foundation for system learning. These types include DenseNet121, DenseNet169, and DenseNet201. The initial stage is to obtain training data from the Indonesian waste database. A total of 1880 data points were trained across 40 epochs using the DenseNet technique. A testing accuracy of 92% was found in the DenseNet121 data. Meanwhile, DenseNet169 training yielded results of 88%. Finally, DenseNet201 has a 92.7% accuracy rate. According to Mao's trash categorization research [30], the DensNet201 algorithm has the highest accuracy level. The high accuracy results are attributable to the network's depth level on DenseNet201. Several misclassifications in the learning model were also investigated in this study. Misclassification images for all three models are shown in Fig. 7.

Conclusion
Artificial intelligence-based mechatronics systems have been developed effectively. Furthermore, Indonesia's waste database has been clearly classified into six sorts. DenseNet201 was used to identify artificial intelligence with the best accuracy, which was 92.7%. In general, mechatronics robot systems based on artificial intelligence may be deployed fully and appropriately. The findings of this study can be used in future research to create a more comprehensive mechatronics system with a full database.