The benefits of customised optical filters for machine vision

The precise spectral control of the filter profile is of crucial importance for machine vision applications and is often the key to the success of complex inspection tasks. In automated quality control, for example, subtle colour differences or material defects often need to be reliably detected, even under difficult lighting conditions or at high process speeds. Interference filters enable precise adaptation to the spectral characteristics of the features to be inspected. They can be designed in such a way that they transmit precisely the wavelength ranges that are relevant for the respective inspection task, while effectively suppressing interfering ambient light.

The importance of this technology becomes particularly clear when looking at specific application examples. In the semiconductor industry, for example, the smallest defects on wafer surfaces must be recognised, whereby different materials and structures must be reliably differentiated. Special interference filters enable the selective observation of certain spectral ranges in which the defects to be recognised stand out particularly clearly. The same applies to the inspection of OLED displays, where different material layers and their quality must be assessed.

Interference filters are manufactured under strict clean room conditions using physical vapour deposition (PVD) in a high vacuum. Modern coating systems utilise various technologies such as electron beam evaporation or sputtering, often in combination with ion-assisted deposition (IAD) processes, which produce particularly dense and stable layers. This stability is an important prerequisite for continuous industrial use in machine vision systems, where the filters are often exposed to extreme environmental conditions.

The process is continuously controlled by high-precision optical monitoring systems that monitor the layer thicknesses in real time as they grow. Modern systems use broadband monitoring systems that record the entire spectrum during the coating process and thus enable optimum process control. Control is carried out using complex algorithms that can immediately recognise and compensate for deviations from the target value.
In addition to process monitoring during coating, all filters undergo extensive quality controls. These include spectrophotometric measurements at various angles of incidence, environmental tests to check temperature and moisture resistance and mechanical tests to ensure coating adhesion and abrasion resistance.

Compared to coloured glass filters, interference filters offer decisive advantages for machine vision applications, which have a direct impact on the performance of the overall system. The steep edges between the passband and stopband enable extremely selective spectral filtering, which is particularly important when differentiating between similar colour tones or detecting specific material properties. While coloured glass filters typically have wide transition ranges, interference filters can achieve slopes with gradients of less than 1% of the central wavelength.

The high transmission in the passband of over 95% ensures efficient utilisation of the available light, which is particularly important for fast inspection processes or weak signals. At the same time, an optical density of OD6 or higher can be achieved in the blocking range, which corresponds to a transmission of less than 0.0001%. This combination of high transmission and excellent blocking is not achievable with conventional filters.

The physical properties of interference filters are optimised for the requirements of industrial image processing. The low temperature dependence of the optical properties – typically less than 0.004 nm/K wavelength shift – ensures stable measurement conditions even at fluctuating ambient temperatures. The thin construction with a total layer thickness of usually less than 50 µm enables compact optical designs, and the high mechanical stability of modern coatings ensures a long service life in continuous industrial use.

Another important aspect is the dependence of the spectral properties on the angle of incidence of the light. Modern filter designs can be optimised in such a way that they largely retain their spectral properties even at larger angles of up to 30° or more. This is particularly important for applications with a large field of view or when using fast lenses.

The development process for customised filters for machine vision begins with a detailed analysis of the application. In close cooperation with the customer, all relevant parameters are recorded: the spectral properties of the objects to be analysed, the characteristics of the lighting, the requirements of the cameras used, but also practical aspects such as design, environmental conditions and cost framework.

Based on this information, the optical design is carried out using specialised software. Various designs are simulated and compared in terms of their performance. Modern optimisation algorithms enable the development of complex coatings that meet all the required spectral properties. Particular attention is paid to the manufacturability and reproducibility of the designs. Suitable tolerance analyses ensure that the filters also meet the required specifications under real production conditions.

After the design phase, prototypes are produced and extensively characterised. The measurement results are discussed with the customer and the design is further optimised if necessary. Series production only begins once all specifications have been met. Strict quality controls ensure consistently high product quality.

The integration of artificial intelligence into machine vision systems places new and specific demands on optical filtering. The quality and specificity of the spectral data have a direct influence on the performance of the AI algorithms. The following principle applies: the better the optical pre-processing, the more reliable the automatic classification and error detection.

Customer-specific interference filters can be specifically optimised here to highlight the spectral features relevant to the respective classification task. By analysing large data sets with machine learning methods, the spectral areas that have the highest discriminatory power for certain defect types or material classes can be identified. The filter designs are then adapted so that precisely these areas are optimally captured.

One particularly interesting aspect is the combination of multispectral imaging with deep learning algorithms. Here, complex filter combinations can be developed that optimise different spectral channels for subsequent AI processing. The high spectral selectivity of the interference filters enables a significant reduction in the amount of data while maximising the information content.

In practice, it has been shown that specially developed filter sets can significantly improve the recognition rates of AI systems. For example, in the automatic quality control of food, the detection rate of foreign objects was increased by more than 30% through optimised spectral filtering. Similar improvements have been achieved in the detection of material defects in semiconductor production.