Su Chang

Mr. Su Chang

orcid

Research Focus & Metrics

Academic Career Length

6Years (2021-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
28 July 2024 in Micromachines
Verified
Journal Article
Open access
PDF file
YLYuhui Liu
Yuhui Liu
DSDuansen Shangguan
Duansen Shangguan
LCLiping Chen
Liping Chen
SCSu Chang
Su Chang
Jing LiuJing Liu
Jing Liu
Investigating the optimal laser processing parameters for industrial purposes can be time-consuming. Moreover, an exact analytic model for this purpose has not yet been developed due to the complex mechanisms of laser processing. The main goal of this study was the development of a backpropagation neural network (BPNN) with a grey wolf optimization (GWO) algorithm for the quick and accurate prediction of multi-input laser etching parameters (energy, scanning velocity, and number of exposures) and multioutput surface characteristics (depth and width), as well as to assist engineers by reducing the time and energy require for the optimization process. The Keras application programming interface (API) Python library was used to develop a GWO-BPNN model for predictions of laser etching parameters. The experimental data were obtained by adopting a 30 W laser source. The GWO-BPNN model was trained and validated on experimental data including the laser processing parameters and the etching characterization results. The R2 score, mean absolute error (MAE), and mean squared error (MSE) were examined to evaluate the prediction precision of the model. The results showed that the GWO-BPNN model exhibited excellent accuracy in predicting all properties, with an R2 value higher than 0.90.

25 June 2024 in Journal of Applied Physics
Verified
Journal Article
Open access
PDF file
DSDuansen Shangguan
Duansen Shangguan
YLYuhui Liu
Yuhui Liu
LCLiping Chen
Liping Chen
SCSu Chang
Su Chang
Jing LiuJing Liu
Jing Liu
Quartz material irradiated by femtosecond laser has increasingly attracted widespread attention for the micro-fabrication of photonic devices. Mechanism exploration is beneficial for accelerating the digital progress of laser processing. However, the mechanism between femtosecond laser and quartz is complicated and needs further theoretical investigation. This paper established the theoretical model based on the ionization model with the Drude equation to study the space–time evolution of free electron density and its influence on the absorption coefficient, reflectivity, and ablation depth. In addition, we achieved a 10 × 10 micro-holes array with a pore size less than 10 μm, cone angle less than 2° in a 0.25 mm thick quartz on the condition of a laser pulse energy Ep = 3 μJ, scanning velocity v = 0.1 mm/s, and defocusing distance Δf = −0.3 mm via the bottom-up femtosecond laser processing. The work gives a new insight into further understanding the ablation mechanism of transparent materials etching by the femtosecond laser. It provides a practical technical scheme for preparing commercial quartz photonic devices.

3 March 2022 in The International Journal of Advanced Manufacturing Technology
Verified
Journal Article
Closed access
PDF file
SCSu Chang
Su Chang
JXJianfeng Xu
Jianfeng Xu

24 March 2022 in Sensors
Verified
Journal Article
Open access
PDF file
SCSu Chang
Su Chang
JXJianfeng Xu
Jianfeng Xu
The case of dual manipulators with shared workspace, asynchronous manufacturing tasks, and independent objects is named a dual manipulator cooperative manufacturing system, which requires collision-free path planning as a vital issue in terms of safety and efficiency. This paper combines the mathematical modeling method with the time sampling method in the classification of robot path-planning algorithms. Through this attempt we can achieve an optimal local search path during each sampling period interval. Our strategy is to build the corresponding non-linear optimization functions set based on the motion characteristics of the dual manipulator system. In this way, the path-planning problem can be turned into a purely mathematical problem of solving the non-linear optimization programming equations set. The spatial geometric analysis is used to linearize the predicted dual-manipulator minimum distance equation, thus linearizing the non-linear optimization equations set. Finally, this system of linear optimization equations will be mapped directly into a virtual Euclidean space and then solved intuitively using the spatial geometry theory. By simulation and comparing with the previous strategies, we find that the planning results of the newly proposed planning strategy are smoother and have shorter deviations as well as a higher algorithmic efficiency in terms of spatial geometric properties.

2 April 2021 in Research Square
Verified
Preprint
Open access
PDF file
SCSu Chang
Su Chang
XJXu Jianfeng
Xu Jianfeng
Analogy to the definition of Human-robot Interaction (HRI), the case of multiple manipulators with shared workspace, non-simultaneous manufacturing tasks and separate objects is named as multi-manipulator cooperation, which is becoming more widely employed in modern industrial manufacturing system and requiring non-collision path planning as a key issue in terms of safety and efficiency. In this paper, a novel method called Sampling based Position Space Map Search (SbPSMS) method which combines the map search method with the time-sampling based method will be proposed, including a minimum distance prediction method based on PSO-BP neural network for collision detection and two candidate position determination methods for search map establishing of all manipulators. After the specific search map simplification process, the local path fragments during each sampling time interval can be determined via cost function, which will be glued together to generate the final collision-free paths. The simulation results not only show that the PSO-BP hybrid algorithm has more accurate of nearly 2mm than the standard BP neural network in minimum distance prediction, but also demonstrated that our proposal can successfully achieve collision avoidance of dual manipulators system whilst meeting the real-time requirements for multi-manipulator cooperate assembling scenarios. The further satisfactory simulation results of triple manipulators suggesting our algorithm can be extended to applications of multiple manipulators cooperate manufacturing.

16 June 2023
Verified
Conference Paper
Closed access
SCSu Chang
Su Chang
SDShengnan Deng
Shengnan Deng
ZHZhiqiang He
Zhiqiang He
JTJian Tan
Jian Tan
GLGuojin Liu
Guojin Liu
Load prediction play a significant role in the field of gas development. In order to effectively mine historical data information and improve the accuracy of short-term gas load, this paper focuses on the gas load based on the time series nonlinear characteristics. A Convolutional Neural Networks (CNN) and Long Short-Term Memory Network (LSTM) are combined to predict the short term load of gas with the advantages of the two types of network. Firstly, CNN is used to extract the important local spatial features of gas load information with 2 layers convolutional network, Then, the latent features of gas load extracted by CNN are feed into the LSTM, which can maintain a good memory of the time series with a long time span, and can fully consider the time correlation of data. Finally, the prediction value of the mixed model is generated according to the deep features of the training data. The simulation result investigates that compared with the CNN, LSTM, BiLSTM and Seq2Seq neural network model, the CNN-LSTM network model performs better to extract the spatiotemporal feature, which can predict short term gas load more accurately.