1. Introduction
Assembly is one of the important and indispensable technologies for the manufacture of products. For the purpose of greater functional integration, the miniaturization of products is inevitable, and the parts to be assembled tend to be miniaturized and diversified, especially with the development of MEMS devices. The assembly accuracy is necessary to be in the range of a few micrometers. Given the limitation of human workers, developing sensor guided or sensor based automatic assembly technology is the solution to achieve high accuracy in a repeatable assembly process [
1]. Among the sensors, the visual image sensor or machine vision system is undoubtedly the most important one, while other sensors including force and proximity are also utilized in assembly application.
Since the 1990s, many research efforts have been conducted on assembly methods and the development of assembly systems for micro devices or MEMS. As there are a variety of different types of small devices, various systems or machines have been developed for the assembly of different miniaturized parts. The use of vision-based feedback has been identified as one of the most promising approaches for controlling the micro assembly process [
2]. As the parts to be assembled are small and specific, the task oriented vision systems utilized in most of the developed micro assembly systems have used microscopic vision with a fixed, small field of view. Some typical examples can be found in references [
3,
4,
5,
6]. When there is a requirement for vision systems with different fields of view, two or multiple vision units with different optical magnifications were employed.
As the automation of precision assembly moves forward, more and more automatic assembly machines are needed for miniaturized device manufacture. For industrial applications, vision systems are required to be more flexible for accomplishing the measurement of parts from the micro to meso scale or even larger.
The precision miniaturized devices manufactured in small and medium-sized enterprises belong to a certain type or category of products with different models, such as micro inertial sensors, bio-MEMS devices or optical modules, et al. The devices consist of a certain number of parts or components to be assembled. However, these parts or components may vary greatly in shape or size due to continuous upgrading or improvement of the devices. Meanwhile, the short R&D cycles result in the manufacturers enduring great pressure and challenges for a quick changeover. It is difficult to fulfill all the assembly tasks of a certain type or category of miniaturized devices with a dedicated assembly system developed for a specific model.
In some dedicated assembly systems developed by our research team, the problem in the application of machine vision due to the limited field of view for high resolution and a large work space was solved with a vision unit mounted on 3 DOF (degree of freedom) precision stages [
7,
8,
9,
10]. However, it is not very efficient in dealing with the parts that have different feature sizes, which may range from several micro meters to tens of millimeters. Furthermore, the adaptability to the device’s changes was limited, and the demands on the flexibility of the assembly system become much more preferred.
In order to develop more versatile assembly systems, which will be capable of accomplishing the assembly tasks of an entire category of miniaturized devices, such as a series of different models of precision accelerometers, and also be adaptable to further revisions, the Modularized Flexible Precision Assembly Station (MFPAS), a modular and more flexible automatic assembly station is being developed at DUT. It has the features of being flexible, expandable and re-configurable to meet wider assembly needs, including adhesive dispensing, micro welding with fiber laser, etc. The MFPAS can adapt to a variety of parts that need to be manipulated or measured. The architecture based on functional modularity has been taken into account in the design phase, which enables the system compactness and re-configuration. The MFPAS is expected to be developed as a modular and flexible system. It adopts the system structure of “Basic platform + functional modules”.
Many scholars have conducted in-depth research on modularity [
11,
12,
13], which has many advantages in both design and manufacturing. Flexibility is an inevitable requirement to respond to product changes. Many efforts have been made in this aspect [
14,
15,
16]. The flexibility of the MFPAS is mainly focused on the module for gripper exchange. A variety of exchangeable grippers already exists [
17,
18,
19]. Unlike the existing ones, the MFPAS is equipped with hollow design structure, which facilitates the machine vision measurement module for viewing through and carrying out measurement tasks from a top view. And the machine vision measurement module can perform multi-scale measurement tasks together with the module for gripper exchange. The gripper exchange brings a high flexibility to the whole system.
With the module for gripper exchange, the system can manipulate a variety of miniaturized parts, and so therefore the system should meet the measurement requirements of the parts with a large size range by taking advantage of the machine vision module.
Compared to the fixed-parameter (passive) lens, a zoom-lens with adjustable focal length has inherent advantages in flexibility and imaging capabilities [
20]. But for precision assembly, the measurement accuracy may be affected due to the change of optical magnification during the zooming process. Thus, an automatic zooming vision system was set up for evaluation and final integration in MFPAS.
The requirements of the machine vision measurement module are listed below:
- (1)
Automatic measuring the target position and orientation of the part (through the hollow structure with no gripper attached) and the position and orientation of the part being grasped (gripper attached).
- (2)
Automatic focusing and amplification adjustment adapting to different parts with the size range of 0.5–10 mm and the size of the finest structure is 5–10 μm, and no image stitch if possible.
- (3)
Measurement accuracy 1.5–5.0 μm with magnification of the lens from high to low, which changes the resolution of the image from high to low; the accuracy for orientation measurement is 0.01° (the length of the part l = 10 mm).
- (4)
No obvious decrease of accuracy with different amplification in single measurement procedure.
- (5)
Automatic calibration during operation when necessary.
In general, a high magnification lens with a small field of view ensures higher measurement accuracy, but makes it difficult to measure parts with larger sizes. When low magnification lenses with large fields of view are used to measure the parts of smaller sizes, it will inevitably reduce the measurement accuracy to some extent. Naturally, the zooming vision system will introduce some new problems, such as inaccurate value of the magnification during zooming procedure. Considering the influence of magnification adjustment on measurement accuracy, pixel equivalent, principal point and orientation deviation of image were analyzed and experimentally studied in this paper. Calibration of pixel equivalent is the precondition and foundation of the measurement based on image. It establishes the proportional relationship between pixels and physical dimensions, and has important influence on positioning accuracy in the XY plane. For camera systems with fixed-parameter lenses, the principal point coordinates are the constituent elements of the camera’s intrinsic parameters, so the principal point can be obtained by camera calibration [
21,
22,
23]. Nevertheless, the principal point could be determined based on the focus-of-expansion method for camera systems with zoom lenses. A new template, circular patterns of different diameters combined with straight lines for the evaluation of the machine vision module, was designed for above parameters’ calibration.
The purpose of this paper is to analyze the factors of the automatic zooming vision, which may affect the measurement accuracy in practical application, and verify the feasibility of the vision system developed for integration in the automatic assembly system MFPAS. In the following section, the overall architecture of the MFPAS is introduced, and the machine vision measurement module and the module for gripper exchange are described briefly. In
Section 3, evaluation of the machine vision measurement module and its experimental testing results are given based on the new template. Finally, the paper is summarized with conclusions in
Section 4.
2. Design of the Modularized Flexible Precision Assembly Station (MFPAS)
The overall design of the MFPAS has the objective of maximizing the flexibility to meet different assembly needs of a series of models of precision miniaturized devices, which belong to a certain type or category of products, including accelerometers and other similar devices. The dimension of the parts is in the range from submillimeters to tens of millimeters, and the assembly accuracy is required to be better than 10 μm. The architecture of “Basic platform + functional modules” is applied to the system, which enables the MFPAS with the ability to quickly minimize the changes or adjustments to meet new assembly requirements.
Basic modules are mainly responsible for ordinary assembly tasks. Additional modules are designed and implemented in the framework of basic modules, and are re-configurable and expandable for special assembly requirements. Thus, the MFPAS can be divided into basic and additional modules. The constituent modules of the MFPAS are shown in
Figure 1.
Figure 2 shows the layout of the MFPAS. The precise 3 DOF positioning module has three degrees of freedom (X, Y, Z) and can realize precise spatial positioning with a large working space (XYZ: 550 × 500 × 300 mm
3). The repeatability of the X, Y axis is 2 μm, and that of the Z axis is 0.5 μm. The positioning error of XY is less than 10 μm, and Z is less than 5 μm. Since the machine vision measurement module and the base mechanism for gripper exchange are both mounted on the Z axis, the measurement task can be achieved after miniaturized parts are picked up by the module for gripper exchange. An assembly force measurement module is integrated in the base mechanism, which is mainly used for contact force control in the Z direction. 1-DOF rotary stage can realize orientation adjustment of parts to be assembled and mounting of fixtures.
2.1. Machine Vision Measurement Module
The machine vision measurement module can achieve accurate measurement of the position and orientation of parts with a large size range from a top view. As shown in
Figure 3, it mainly consists of a CCD camera, zoom lens and light sources. Some technical parameters of the CCD camera are listed in
Table 1.
The selected zoom lens is NAVITAR ZOOM6000 (Navitar, Inc, New York, NY, USA), which is a telecentric lens used to eliminate parallax between an image and the object in measurement. The zoom is a stepless type, which is continuously controlled by a motor, and the range of magnification is 0.7×–4.5×. The change in focal length of the lens allows the machine vision measurement module to have the ability to measure feature sizes from several micrometers to several millimeters, the biggest feature size can be as large as 12.5 × 9.4 mm2. The CCD camera and zoom lens are mounted on the Z axis, which gives it a wide measurement space as large as the working space.
Different miniaturized parts have different assembly accuracy requirements, ranging from a few microns to tens of microns. Besides the positioning errors of the XYZ guide rails and the system installation errors (including guide rail, camera), the measurement errors of the vision system also affect the assembly accuracy and should also be considered. The measurement errors are caused by several factors of the vision system, including pixel equivalent, principal point and orientation deviation of image, which should be confirmed with different magnifications of the lens. This paper focuses on these measurement errors. For large measurement errors, online calibration or a necessary compensation strategy should be taken according to the specific parts assembly accuracy requirements.
2.2. Modules for Gripper Exchange
Modules for gripper exchange of the MFPAS consists of the base mechanism for connection and exchangeable tool magazine. The base mechanism for connection adopts a hollow design that facilitates the observation of the machine vision measurement module through it, as well as simplifies the system structure. As shown in
Figure 4, it is also mounted on the Z axis, which gives it a wide operating space.
An exchangeable tool magazine is shown in
Figure 5, where a different gripper or assembly tool heads are positioned by using the locating pins and holes in the magazine. The exchangeable tool magazine has two degrees of freedom (
x-axis translation and ɑ rotation), and a 3-DOF base mechanism for connection can automatically pick and place different tool heads from the magazine. Thus, it is more flexible to perform various assembly tasks. In addition, multiple grippers or assembly tool head positions are reserved to enhance the expandability.
Different tool heads can be exchanged by using electromagnets for connection and disconnection with the base mechanism. This electromagnet has the advantages of simple control, high reliability and low power consumption. The magnetic force disappears only when the voltage of certain value is supplied, otherwise the magnetic attraction force exists. Therefore, we can easily control the electrical signal to exchange different tool heads. Moreover, the characteristic of this electromagnet ensures that the tool head will not be released even in the case of unexpected power loss.
Figure 6 shows the schematic diagram of picking a tool head by the base mechanism for connection. After a specific tool head is selected, the parts are picked up. The accurate measurement of the position and orientation of the parts can be achieved from a top view via the machine vision measurement module.