Machine vision is constantly being applied in various fields. What challenges does it face?

As one of the important components of industrial automation systems, the technology and application of machine vision are becoming increasingly mature with the development of the automation industry, which is reflected in: the image processing capability and speed are constantly enhanced, the performance of optoelectronic devices is improved, various standards are gradually unified, and the price is relatively reduced.

According to the market research report of AIA (Automated Imaging Association), the global machine vision market size exceeded 7 billion US dollars in 2006, and it is predicted that it will continue to grow in the next five years. However, as suppliers and integrators continue to promote machine vision applications in various fields, how to seamlessly integrate machine vision, a relatively independent function, into various types of automation equipment in various industries has encountered unprecedented challenges.

1. Applications and Challenges of Machine Vision

Machine vision applications can be divided into two main categories:

One type is used in large-scale or high-testing production lines, such as packaging, printing, sorting, etc., or in environments that are not suitable for human work, such as the field and nuclear power plants, using machine vision to replace traditional manual measurement or inspection, while achieving reliability, accuracy and automation that cannot be achieved under manual conditions.

Another type of application is the manufacturing of professional equipment that must use high-performance, precision machine vision components. The typical representative is the semiconductor manufacturing equipment that first led to the rise of the entire machine vision industry. From the upstream wafer processing and manufacturing classification and cutting to the end circuit board printing and patching, this type of equipment relies on high-precision visual measurement to guide and position moving parts. For example, if there is a positioning deviation in the solder paste printing process, and the problem is not discovered until the online test after the chip is mounted, the cost of rework will be more than 100 times the original cost.

However, in the above applications, machine vision functions are rarely used as isolated systems, but as an organic part of the entire automation system or equipment. They can only truly play their advantages when combined with other functions such as logic control, motion control, data acquisition, communication networks, and enterprise database management. In addition to completing a series of processes from light source allocation to image processing software development, building a machine vision system also faces the challenges of integrating with the above-mentioned complex automation system functions. A single vision development software and hardware solution often means that the overall development cycle, cost, and uncertainty risks of the automation system must be borne by the manufacturer or integrator. The difficulty of integrating machine vision with automation systems has largely hindered its application in the relatively conservative field of industrial automation.

2. Solutions based on NI LabVIEW and machine vision system

Faced with the above challenges, the NI LabVIEW software platform and its machine vision system provide a good solution.

Let's first look at the development and integration process of machine vision from the perspective of software: First, with the help of efficient and convenient configuration software VBAI (machine vision generator for automatic detection) and comprehensive vision modules (covering support for all formats and standards of cameras, providing pattern matching, OCR, particle analysis, 2D barcode recognition and hundreds of other image processing functions), users can verify different camera and light source settings, acquisition methods and image processing algorithms in an interactive development environment, and then automatically generate executable programs corresponding to LabVIEW for the confirmed steps. The LabVIEW software platform has intuitive graphical development features, allowing engineers to focus more on function development rather than code writing.

In the overall system development and integration process, engineers can directly use the corresponding LabVIEW toolkits and modules to complete motion control, data acquisition, industrial communication and human-machine interface functions in the same way under a unified platform, and achieve connection and communication with various PACs (programmable automation controllers), PLCs, industrial equipment, OPC clients and corporate databases. For this development model, both experienced integrators and junior developers are freed from the difficulties of dedicated or even private development methods and platforms, drivers and protocols, and physical communication and synchronization between devices corresponding to different devices, which greatly reduces the difficulty and cost of system integration.

From the perspective of hardware architecture, PC-based machine vision systems, due to their openness and flexibility, provide powerful processing capabilities while being easier to integrate with other functions. However, due to reasons such as reliability and size, the PC architecture cannot fully meet the needs of industrial applications.

Another way is embedded architecture, which is easy to use and highly reliable, but has relatively simple functions and poor integration. In order to solve these contradictions, NI has integrated LabVIEW real-time and FPGA technology in its compact machine vision system (CVS), and has unprecedentedly realized the flexible customization of I/O and communication protocols and motion on the same embedded hardware platform. It can simultaneously collect and process 3-way image signals, and ensure the robustness and reliability of the system to meet the application requirements in the harsh environment of industrial sites (Figure 1).

FIG.1 FIG.1 NI compact machine vision system

Below we will analyze two examples to explore in detail how to use open and flexible hardware and software platforms to integrate machine vision and multi-domain functional applications to reduce system integration complexity and shorten development cycles.

3. Automated semiconductor wafer sorting system based on LabVIEW and synchronized machine vision, motion control, and data acquisition

In the semiconductor manufacturing industry, wafers must be carefully classified according to electrical and physical parameters such as thickness (THK), total thickness tolerance (TTV), bow (BOW), and warpage (WARP) before cutting to meet strict tolerance requirements. To ensure measurement accuracy, the traditional single-point measurement method consumes a lot of test time.

To this end, Gigamat Technologies of the United States has developed a new generation of full-scan automatic sorting equipment (Figure 2) to improve throughput and meet the accuracy and repeatability requirements under single-point testing, which is a considerable challenge in technology.

Fig.2 Automated semiconductor wafer sorting system 

The new fully automatic wafer sorting system makes full use of the LabVIEW platform and its supporting toolkit. The system is divided into two working steps: wafer alignment and measurement. The alignment process uses line scan image acquisition and 3-axis motion control. By synchronizing image acquisition with chassis rotation rate, the image acquisition of 6 million pixels of the entire wafer is completed within 1 second. The LabVIEW visual algorithm is used to determine the center position, flatness and other characteristics of the wafer, and the wafer position is adjusted accordingly to achieve full matching with the parameter measurement platform.

The measurement procedure requires that the resolution of the distance between the upper and lower surfaces be less than 0.0001mm. The solution is to use the NI motion control tool on the LabVIEW platform to generate a smooth arc and spiral trajectory combination, accurately control the position of the rotating chip, and use the NI data acquisition card to complete multi-channel synchronous high-speed and high-density measurement of the probe, record the corresponding position in real time, perform relevant calculations and processing based on this, obtain various parameter information, and finally obtain the classification result.

In addition to the above core steps, the system also includes: touch screen human-machine interface; wafer lift control based on RS-485 communication; digital I/O control for light source, machine power and vacuum equipment; and connection with Microsoft Access database to realize digital processing of the processing process. These functions are all developed uniformly under the LabVIEW platform. Gigmat's manager commented that "without LabVIEW and the synchronization of NI machine vision, motion control and data acquisition products, this project would not be economically feasible."

NI Compact Machine Vision System Helps Automotive Spark Plug Inspection Achieve Six Sigma Repeatability

The eccentricity and electrode spacing of automotive spark plugs are key indicators that determine their performance. In the past, a leading automotive spark plug manufacturer had been measuring them manually. Due to low measurement accuracy, it was necessary to use overly strict product tolerance limits, resulting in unnecessary production requirements and reduced output. In order to ensure reliable quality control, faster inspection speed and increased output, the manufacturer decided to establish a full-scan dimensional quantification system based on machine vision.

The system consists of an IEEE 1394 camera, a ring light source, a rugged NI CVS embedded machine vision system, and a LabVIEW software development platform. The collected spark plug image is transmitted to the CVS through the fire line, and special algorithms such as real-time circular edge detection are run on it. By controlling undersampling to find the balance between accuracy and processing time, the measurement accuracy reaches 0.01mm, which fully meets the 6Sigma standard.