Zengguo Tian

With the rapid development of microelectronics science and technology, the quality of IC-grade silicon single crystal directly affects the yield and stability of the performance of semiconductor device production. As the main equipment for the preparation of such materials, the monitoring and maintenance of the working condition of the single crystal furnace are crucial. Bi-directional long short-term memory (Bi-LSTM) is an innovative neural network paradigm that is used to predict future occurrences by learning the bi-directional long-term dependencies of time-steps and serial data. This paper built a Bi-LSTM based model that can dynamically predict the pulling speed of a Czochralski (Cz) single-crystal furnace by modeling the time series of operational parameters. The Bi-LSTM model is validated using real data from a silicon single-crystal factory. It is proven that the model achieved higher accuracy than LSTM, ANN, SVR, and XGBOOST. The experimental results verify the validity of modeling the pulling speed of single-crystal furnace devices through the Bi-LSTM model by using the time series of multi-dimensional parameters. Therefore, the Bi-LSTM model can serve as a reference for modeling the parameters of such devices.

To solve the problem in which the output power and wavelength of semiconductor lasers in fiber optic sensing systems are easily affected by the drive current and temperature, a high-precision current drive and temperature control system was developed in this study. The embedded system was used to provide a stable drive current for the semiconductor laser through closed-loop negative feedback control; moreover, some measures, such as linear slow-start, current-limiting protection, and electrostatic protection, were adopted to ensure the stability and safety of the laser’s operation. A mathematical model of the temperature control system was constructed using mechanism analysis, and model identification was completed using the M sequence and differential evolution (DE) algorithms. Finally, the control rules of the fuzzy proportional integral differentiation (PID) algorithm were optimized through system simulation to make it more suitable for the temperature control system designed in this research, and the accurate control of the working temperature of the semiconductor laser was realized. Experimental results showed that the system could achieve a linearly adjustable drive current in the range of 0–100 mA, with an output current accuracy of 0.01 mA and a temperature control accuracy of up to 0.005 °C.

The ultrasonic welding (UW) technique is a fast-joining process; it is very suitable for the carbon fiber reinforced thermoplastic (CFRTP) composite. For improving the consistency of the welded joint quality, a new pre-pressing ring clamp (PPRC) was designed for ultrasonic welding carbon fiber reinforced nylon composites in this paper. The effects of the PPRC on the weld quality of the ultrasonic welding welded 4.0 mm thick 30% mass short carbon fiber reinforced Nylon 6 composite was investigated and compared with that of normal clamp weld joint. The weld strength, microstructure, and temperature evolution of the joint were analyzed by tensile test, scanning electron microscope, and temperature measurement. The results showed that the PPCR UW joints had larger central weld nugget size (478 mm² vs. 300 mm²), thicker stable fusion region thickness (1.10 mm vs. 0.96 mm), resulting in a higher joint strength (6.86 kN vs. 6.21 kN) compared with the normal clamp UW joints under the same welding parameters. The real-time monitor curve of the horn displacement and temperature at the faying interface showed that the PPRC increased the heat rating at the faying interface during instable melting stage. The PPRC could improve the contact condition between workpieces and the utilization efficiency of ultrasonic energy, which boosted the melting rate of materials at faying interface and consequently the formation of a sound joint with enough weld size (i.e., 433 mm²) in a shorter welding time (i.e., 1.3 s). Therefore, the flexibility of component assembly would be increased by the use of this sort of clamps.

Ultrasonic welding (UW) of polymeric composites is significant in automobile industry; however, maintaining the perfect contact condition between workpieces is a great concern. In this study, effect of preloading and welding pressure on strengths of UWed 2.3-mm-thick short carbon fiber reinforced nylon6 (C_f/PA6) joints with poor contact between workpieces was investigated through stress simulation and energy dissipation at the faying interface. Results showed the application of preloading can increase the strength of normal joint by 18.7% under optimal welding parameters. Gaps between upper and lower workpieces decreased the joint strength significantly, especially for gaps greater than 1.5 mm. Preloading improved the strengths of the joints with gaps remarkably, where the strength of joints with 1.5 mm gap recovered to 95.5% of that the normal joint. When combining the weld nugget evolution, stress-deformation simulation during UW, and ultrasonic vibration transmission analysis, the improvement mechanism of the joint under preloading was mainly because the preloading compacted the contact between workpieces, which favored the energy transmission at faying interface.

Wafer identification (ID) is a serial number printed on the surface of wafer, which is used for indexing production process data in manufacture execution system. The automatic recognition of wafer ID is helpful to improve the level of automatic production. However, the existing equipment and methods mainly focus on single wafer-ID recognition, which require wafers to be taken out and placed on a specific platform, resulting in low efficiency. In this paper, we present a batch wafer-ID recognition method based on machine vision, including a specific designed image-acquisition system and recognition algorithms. Based on the priori information, we formulate a 3D model for the cassette and wafers to be registered with the features extracted from the image. Combined with image-processing techniques, the pose of wafers in the cassette is estimated to undistort the perspective deformation of wafer-ID characters, such that we can exploit a classic lightweight convolution neural network for character recognition. The proposed system can capture and recognize the whole image of a cassette of wafers at once, which does not need to take out the wafers from the cassette and avoids the risk of contamination. Extensive experiments were conducted to evaluate the performance of our proposed techniques by collecting the data set of wafer-ID images. The results show that our proposed 3D model-based rectification method can correct the character deformation effectively and enable the lightweight classifier to achieve high speed and high accuracy for batch wafer-ID recognition.

The digital camera was used to collect crop real scene images in automatic observation of crop meteorology, and image processing algorithm was used to extract crop characteristics. Due to the influence of camera distortion parameters, the real images have distortion in different degrees, and there are deviations in the recognition results of the observed area, density and incubation period. According to the results of the crop automatic identification, contrast the distortion parameters of camera was analyzed, and the camera calibration algorithm is studied. Moreover, the accuracy of crop automatic observation was improved, due to the image correction.

This article has designed a vehicle gasoline on-site sampling and rapid detection device. Automatically taking gasoline samples during gasoline handling, two-dimensional gas chromatographic method is used for the rapid quantitative detection of the unconventional additive methylal. The heavy components are separated by back-blowing to air with the non-polar capillary column while the methylal and light components are separated by a polar column to determine the methylal content. The experiment shows that linear relationship between value of peaks area and mass concentration of methylal is keep in the range of 0.020~90.0mg/mL with the correlation coefficients of 0.99, and the relative standard deviation (RSD) are less than 2%. The study shows that the limits of detection are 0.012g/L. This device can provide effective means for vehicle gasoline on-site sampling and quality control.