Research on assembly quality adaptive control system for complex mechanical products assembly process under uncertainty

Highlights

• An assembly quality adaptive control system is proposed, developed and applied.

• The system is utilized to reach monitoring, data capture, tracing, and analysis.

• Assembly quality control threshold optimization is built based on BP neural network.

• Assembly performance prediction is built based on BP neural network.

• The assembly precision, stability and efficiency are improved by applying the system.

Abstract

To improve the products’ assembly precision, stability and efficiency, an assembly quality adaptive control system (A_QACS) is proposed for complex mechanical products under uncertainty. Firstly, the characteristics of the complex mechanical products assembly process are analyzed, and the definition and overall framework of assembly quality adaptive control (AQAC) are elaborated. Then, the operation logic of AQAC is researched based on the definition of six units. And the structure of A_QACS system is presented for complex mechanical products. Under the structure, the key technologies of the A_QACS system are described respectively, including assembly resource identification, the key source of monitoring, the data acquisition technology and assembly quality adaptive optimization technology. The monitoring of assembly process, real-time data acquisition, assembly quality adaptive control as well as closed-loop intelligent control integrated with comprehensive quality information tracing and knowledge accumulation are realized. It provides support for timely optimizing the work instructions and control strategies for operators and equipment. Finally, an example of the selection of main bearing and the assembly of crankshaft is given. The system is demonstrated feasibility and effectiveness by analyzing variance, assembly qualification rate and rework rate. And the proposed system provides a practical and efficient technology for assembly process quality control.

Introduction

The assembly precision, stability and efficiency of products have become an important guarantee for enterprises to adapt to market demand [1], [2]. The assembly process quality control is the key to ensure the quality of the products. However, the products assembly process is accompanied with the flow of material, information and error. And the defects may be propagated, accumulated and amplified with WIP (work-in-process) to downstream process. So the assembly quality of products may have a large fluctuation [3], [4]. The defect of the assembly process will seriously affect the quality and cost of the final product. It is difficult to control the assembly quality with considering many uncertain factors, and the relationship between them is non-linear and traditional method is hard to solve it. Hence, the assembly process adaptive control is an effective method to improve products quality. Considering the coupling relationship between the downstream and upstream quality control points of assembly process, the assembly quality optimization and the assembly performance prediction are important problems in the assembly process adaptive control for complex mechanical products. Radio frequency identification technology (RFID) provides the possibility for the assembly quality control [5], [20], [21], [24], [25], [27], [28].

The related work of quality control can be divided into three categories: assembly quality optimization, assembly performance prediction and quality control system, as show in Table 1.

Liu et al. [6] proposed the assembly cost–tolerance model, on the basis of researching on the relationship between the assembly tolerance and the cost. For gaining the minimum assembly cost, the online tolerance optimization model was structured based on shortest path. Cheng and Tsai [7] established cost–tolerance model with minimum manufacturing cost as the target and tolerance standards and equipment processing capacity as the constraints, and optimization of the model using the Lagrange multiplication and Lambert W function. Considering the impact of uncertain factors on quality control and quality assurance in assembly process for mechanical products, the assembly process model, the activity control model and the quality data model were proposed and detailed to realize the product assembly process quality control [8]. In addition to considering the manufacturing cost and the quality loss, Huang and Shiau [9] improved the tolerance optimal allocation model by introducing the reliability index. To improve the assembly quality of product, Karmakar and Maiti [10] proposed the issues of tolerance synthesis, such as construction of design function, construction of the objective function, and selection of suitable optimization methods.

Improper design, defective part, variance in assembly system, and operator error are the main causes of assembly defects [11]. Su et al. [11], [12] presented a novel defect-rate prediction model that is derived from the study of the design-based assembly complexity factor and the process-based assembly complexity factor, which are defined according to the structure and production characteristics of the product. Considering that the dimension and geometric tolerance consists of three classes, namely deterministic constraint, micro-degree constraint, and release constraint, Liu et al. [13] proposed a uniform model called maximum compatible constraint model to solve the problem of precision predicting for complicated product. To predict the assembly process reliability, Zhang et al. [14] established the relationship model between the data of key reliability control point and the assembly process reliability with grey system theory based on determining reliability control points. Xu et al. [15] proposed a comparative study of the predictive performances using the neural network time series models. Hong et al. [16] built a generic mathematical model and state space model to predict and control the variation in assembly process and discover the assembly process effect on final accuracy. Aiming at the problem of assembly precision prediction for spacecraft, Hu et al. [17] constructed a prediction model of assembly accuracy, which consists of an assembly model, a part model, a structure model and tolerance model based on 3D model. To gain the competitive advantage in manufacturing companies, Du et al. [18] proposed a stream of variation (SoV) methodology developed and applied. And the methodology could realize analysis, prediction and control of product quality and productivity improvement in complex multistage manufacturing systems.

Future manufacturing systems need to cope with frequent changes and disturbances. As such, their control requires constant adaptation and high flexibility [19], [20]. Attention is to design algorithms for effective and automated conflict and error prevention and detection in complex networks [21]. So, the point is to build an effective quality information system that can realize assembly process monitoring, data acquisition, process information tracing and querying, etc. Mahdavi et al. [22] proposed a real-time quality control information system that could help control the flow of quality-related information in the production network and ensure that it was stored to be accessible whenever needed at a later time for a variety of uses. To apply well to manage the distributed manufacturing in a multi-company, Helo et al. [31] proposed the core of architecture for next MES solution and developed a pilot software tool to support the needs related to real time, cloud-based, light weight operation.

It is very important to provide the right information to the right person at the right time in the right format to achieve optimal production control in the manufacturing process and to track the real-time manufacturing information [23]. Zhang et al. [23], [24] researched the application of RFID technology in the manufacturing process for the real-time data acquisition and manufacturing process monitoring, such as real-time person information, manufacturing cost, WIP, inventory, etc. In order to better solve the problem of production line stop with orders change, material shortages and equipment failure, Rolón and Martínez [25], [26] defined an intelligent manufacturing execution system realizing the monitoring and analysis of order Agent and resource Agent. And the system has been developed to allow selfish behavior and adaptive decision-making in distributed execution control and emergent scheduling [25], [26]. Zhong et al. [27] presented an RFID-enabled real-time manufacturing execution system to adapt the shop-floor uncertainty and complexity. And RFID devices were deployed systematically on the shop-floor to track and trace manufacturing objects and capture real-time production data. W.N. Liu et al. [28] designed a production management system integrated with an RFID-enabled real-time data acquisition system for motorcycle assembly line. And the system increased the visibility and traceability of assembly process and could cope with the complex of assembly process. To satisfy the predetermined assembly tolerance, Iyama et al. [29] considered the part flow in a high-quality relay production system applying the corrective assembly approach incorporating machining, measurement and reprocessing errors simultaneously, and formulated the production rate of high-quality products. Kwon et al. [30] presented an advanced process management method called Procedure Tree (PT). And the PT could manage massive RFID data effectively and perform real time process management efficiently. A unified approach to the design, planning and control of mechanical and electronic assembly systems was present [42].

Above all, the issues of quality optimization, control and prediction are attracting scholars of the world in recent years. And their researches bring excellent effects in practice. The results of those researches provide the basis for the assembly quality adaptive control. However, the assembly process has high uncertainty and abnormal fluctuation. So, the assembly quality control needs to be integrated with the real-time assembly process. Then, operators or equipment can be timely and accurate dealing with various kinds of abnormal fluctuation. However, there are relatively fewer studies about assembly quality adaptive control for complex mechanical products in assembly process under uncertainty. Besides, the assembly performance of products is decided by the various quality characteristics, such as dimensional tolerance, torque, displacement, flatness, roughness, etc. However, few literatures focused on the assembly process quality control with considering the various types of quality characteristics and the relationship between them. So, the purpose of this paper is to build an assembly quality adaptive control system (A_QACS) for complex mechanical products assembly process under uncertainty based on the previous studies.

The system takes assembly precision, stability and efficiency as the targets. Combined with the characteristics of complex mechanical product assembly process, an assembly quality control method is proposed based on particle swarm optimization of BP neural network (PSO-BPNN). The assembly quality adaptive control (AQAC) includes the assembly quality control threshold optimization (AQCTO) and the assembly performance prediction (APP) based on PSO-BPNN. The assembly results of the upstream process input to the AQCTO model. So, the current process can get the optimal assembly solutions. After completing several assembly processes, the performance parameters of the WIP input the APP model. And the control strategies are given by the prediction results. Finally, the A_QACS system is developed and applied in an engine assembly line.

This paper is organized as follows. First, the characteristics of complex mechanical products’ assembly process and the overall framework of assembly quality adaptive control are presented (in Section 2). Section 3 describes the operation logic of assembly quality adaptive control and the structure of A_QACS. The key enabling technologies for A_QACS systems are introduced in Section 4, mainly including assembly resource identification, the key source of monitoring and data acquisition technology, assembly quality adaptive optimization technology based on BP neural network. In Section 5, a motivating case study is presented to show the effectiveness of A_QACS. Some conclusions and remarks are noted in the final section.