This article presents an online learning method for improved control of nonlinear systems by combining deep learning and fuzzy logic. Given the ability of deep learning to generalize knowledge from ...training samples, the proposed method requires minimum amount of information about the system to be controlled. However, in robotics, particularly in aerial robotics where the operating conditions may vary, online learning is required. In this article, fuzzy logic is preferred to provide supervising feedback to the deep model for adapting to variations in the system dynamics as well as new operational conditions. The learning method is divided into two phases: offline pretraining and online posttraining. In the former, the system is controlled by a conventional controller and a deep fuzzy neural network (DFNN) is pretrained based on the recorded input-output dataset, in order to approximate the inverse dynamical model of the system. In the latter, only the pretrained DFNN is used to control the system. In this phase, the fuzzy logic, which encodes the expert knowledge, is utilized to observe the behavior of the system and to correct the action of DFNN instantaneously. The experimental results show that the proposed online learning-based approach improves the trajectory tracking performance of the unmanned aerial vehicle.
In wire and arc additive manufacturing (WAAM), it is crucial to maintain a constant travel speed in order to achieve a uniform bead morphology along the tool-path. However, this requirement is always ...violated in practice because the welding robot has to slow down when sharp corners with high curvatures are encountered so as to satisfy the dynamic constraint. The consequence is excessive fillings (i.e. humps) around these sharp corners, which not only increase the required post processing, but also prevent the continuation of the deposition process if the accumulated errors in the build direction become too large. This issue greatly limits the application of WAAM in the fabrication of complex-shaped components. In this paper, an adaptive process control scheme (APCS) capable of guaranteeing a uniform bead morphology while still respecting the dynamic constraint is proposed. First, the APCS divides the tool-path into several segments depending on whether they contain sharp corners. Then, for each segment, the APCS automatically selects the allowable travel speed subjected to the dynamic constraint, and also the wire-feed rate according to a process model established in advance. Through the matching between the travel speed and the wire-feed rate, a uniform bead morphology among different segments is achieved. Experiments were conducted on a gantry robot using the Cold Metal Transfer (CMT) process, controlled by a self-developed computer numerical control (CNC) system, validating the efficacy of the proposed scheme.
Injection molding is a cyclic process of producing plastics products by injecting molten plastic materials. There is much complexity in the process because it is a highly non-linear process; many ...factors affect the product quality after long-run fabrication processes. In this study, product weight is defined as product quality, and this research aims to minimize the variation in the product weight. In this study, a standard process parameter setup procedure is proposed in order to set the process parameters in a close to optimized value based on the information collected by the sensors so the adaptive process control system can work well. Three polypropylene materials with low to high MI (melting index) values were used to verify the adaptive control system. The study with the information collected through two sensors installed in the machine—the injection nozzle pressure sensor and tie-bar strain gauge. Based on a series of results in the preliminary experiments, the appropriate V/P switchover point and clamping force could be judged from the peak pressure and clamping force peak for materials with different viscosities, respectively. The viscosity index and peak pressure strongly correlated with the weight of the injection molded parts toward the materials of different viscosity. The viscosity index, peak pressure, and clamping force peak are defined in this research as the quality monitoring characteristics used to guide the adjustment and increase the stability of the process in the presented control system. The experiment reveals that a combination of nozzle pressure curve-based adaptive process control incorporated with clamping force peak monitoring can decrease the weight variation of the injection molded parts. With this system, the coefficient of variation of product weight (Cv) of different viscosity materials from low to high can be decreased to 0.031%, 0.020%, and 0.024%, respectively.
•The viscosity index, peak pressure, and clamping force peak are defined in this research as the quality monitoring characteristics used to guide the adjustment and increase the stability of process in the presented control system.•The experiment reveals that a combination of nozzle pressure curve based adaptive process control incorporating with clamping force peak monitoring can decrease the weight variation of the injection molded parts.
This work presents an in-situ quality assessment and improvement technique using point cloud and AI for data processing and smart decision making in Additive Manufacturing (AM) fabrication to improve ...the quality and accuracy of fabricated artifacts. The top surface point-cloud containing top surface geometry and quality information is pre-processed and passed to an improved deep Hybrid Convolutional Auto-Encoder decoder (HCAE) model used to statistically describe the artifact's quality. The HCAE’s output is comprised of 9 × 9 segments, each including four channels with the segment's probability to contain one of four labels, Under-printed, Normally-printed, Over-printed, or Empty region. This data structure plays a significant role in command generation for fabrication process optimization. The HCAE’s accuracy and repeatability were measured by a multi-label multi-output metric developed in this study. The HCAE’s results are used to perform a real-time process adjustment by manipulating the future layer's fabrication through the G-code modification. By adjusting the machine's print speed and feed-rate, the controller exploits the subsequent layer’s deposition, grid-by-grid. The algorithm is then tested with two defective process plans: severe under-extrusion and over-extrusion conditions. Both test artifacts' quality advanced significantly and converged to an acceptable state by four iterations.
Injection molding is a highly nonlinear procedure that is easily influenced by various external factors, thereby affecting the stability of the product's quality. High-speed injection molding is ...required for production due to the rapid cooling characteristics of thin-walled parts, leading to increased manufacturing complexity. Consequently, establishing appropriate process parameters for maintaining quality stability in long-term production is challenging. This study selected a hot runner mold with a thin wall fitted with two external sensors, a nozzle pressure sensor and a tie-bar strain gauge, to collect data regarding the nozzle peak pressure, the timing of peak pressure, the viscosity index, and the clamping force difference value. The product weight was defined as the quality indicator, and a standardized parameter optimization process was constructed, including injection speed,
/
switchover point, packing, and clamping force. Finally, the optimized process parameters were applied to the adaptive process control experiments using the developed control system operated within the micro-controller unit (MCU). The results revealed that the control system effectively stabilized the product weight variation and standard deviation of 0.677% and 0.0178 g, respectively.
To increase the productivity of injection molding machines, we developed a smart injection part weight stability control system based on C++ programming and domain knowledge. The proposed system is ...meant to eliminate variability in the quality of injected parts by adjusting the changeover position. We developed a viscosity index based on melt pressure data related to guide the adjustment to the changeover position in accordance with material properties. This was achieved by mounting a pressure sensor on the nozzle of the injection molding machine to enable the on-line monitoring of pressure throughout the injection molding process. A series of experiments was conducted to characterize the relationship between viscosity index and injection-molded samples in order to validate the efficacy of the proposed injection stability system. Single-factor experiments were conducted with the changeover position and melt temperature as parameters. The quality of the molded samples obtained under different process parameters was evaluated in terms of weight. Experiment results revealed a correlation between changes in viscosity index and changes in the weight of the samples. The injection stability system can also be operated in self-adjusting mode, in which the changeover position is varied according to viscosity index. In experiments, abnormal machine operations prompted the adjustment of changeover position. Variation in the weight of parts was used to define an index to validate the efficacy of the proposed system.
This research developed an adaptive control system for injection molding process. The purpose of this control system is to adaptively maintain the consistency of product quality by minimize the mass ...variation of injection molded parts. The adaptive control system works with the information collected through two sensors installed in the machine only-the injection nozzle pressure sensor and the temperature sensor. In this research, preliminary experiments are purposed to find master pressure curve that relates to product quality. Viscosity index, peak pressure, and timing of the peak pressure are used to characterize the pressure curve. The correlation between product quality and parameters such as switchover position and injection speed were used to produce a training data for back propagation neural network (BPNN) to compute weight and bias which are applied on the adaptive control system. By using this system, the variation of part weight is maintained to be as low as 0.14%.
The injection-molding process is a non-linear process, and the product quality and long-term production stability are affected by several factors. To stabilize the product quality effected by these ...factors, this research establishes a standard process parameter setup procedure and an adaptive process control system based on the data collected by a nozzle pressure sensor and a tie-bar strain gauge to achieve this goal. In this research, process parameters such as the
switchover point, injection speed, packing pressure, and clamping force are sequentially optimized based on the characteristics of the pressure profile. After the optimization process, this research defines the standard quality characteristics through the optimized process parameters and combines it with the adaptive process control system in order to achieve the purpose of automatic adjustment of the machine and maintain high-quality production. Finally, three different viscosity materials are used to verify the effectiveness of the optimization procedure and the adaptive process control system. With the system, the variation of product weight was reduced to 0.106%, 0.092%, and 0.079%, respectively.
To survive in the highly competitive global economy, manufacturing systems must be able to adapt to new circumstances. An important prerequisite for adaptation is the ability to learn, a process ...based on knowledge discovery and growth. The aim of this research is to uncover knowledge by examining a large volume of real-time manufacturing data collected during manufacturing operations and to use the insights gained to support decision-making and adaptive process control. The paper presents the concept of a self-learning autonomous work system. This concept introduces a learning loop into a manufacturing system composed of data acquisition, data mining (DM), and knowledge-building models. Two methods for DM are applied. A descriptive DM method enables discovery of patterns in data that may contribute to a better understanding of the manufacturing processes. A predictive process provides knowledge in the form of rules, which can then be used for enhanced decision-making. To illustrate the utility of the knowledge models, the concept of adaptive process control is introduced and implemented in a high pressure die-casting domain. A case study based on industrial data collected during die-casting operations provides a demonstration of the concept.
In this study, fed-batch cultures of a
Pichia pastoris
strain constitutively expressing a single chain antibody fragment (scFv) under the control of the glyceraldehyde-3-phosphate dehydrogenase (GAP) ...promoter were performed in a pilot 50 L bioreactor. Due to the very high cell density achieved within the first 75 h, typically between 140 and 160 g-DCW/L of dry cell weight (DCW), most of the scFv is produced under hard oxygen transfer limitation. To improve scFv productivity, a direct adaptive dissolved oxygen (DO)-stat feeding controller that maximizes glycerol feeding under the constraint of available oxygen transfer capacity was developed and applied to this process. The developed adaptive controller enabled to maximize glycerol feeding through the regulation of DO concentration between 3 and 5 % of saturation, thereby improving process productivity. Set-point convergence dynamics are characterized by a fast response upon large perturbations to DO, followed by a slower but very robust convergence in the vicinity of the boundary with almost imperceptible overshoot. Such control performance enabled operating closer to the 0 % boundary for longer periods of time when compared to a traditional proportional–integral–derivative algorithm, which tends to destabilize with increasing cell density.