Establishing automatic coating paths: empirical study for steel component production

Steel is very widely used in construction, whether in the form of steel rebars within reinforced concrete structures or as steel components in steel structures. While steel structures boast numerous advantages such as high strength, low weight, impressive toughness and malleability, they also possess certain shortcomings. Specifically, steel structures exhibit subpar fire resistance and susceptibility to corrosion (Bo et al., 2024). The occurrence of either of these issues can have detrimental effects on the structure, diminishing its lifespan and compromising safety. To mitigate these concerns, surface coating of steel components is of paramount importance (Zhang et al., 2022). This crucial step enhances the ability of a steel structure to withstand the impacts of external environmental factors.

H-shaped steel components are particularly prevalent in construction. They serve not only as beam–column supports in building structures but also as integral beam components in steel bridges (Zhang et al., 2023a, 2023b). Consequently, the primary focus of this study was centred on H-shaped steel components. Currently, the predominant approach for applying surface coatings to these components is a conventional manual method. This method relies heavily on a significant workforce and lacks a standardised spraying approach, leading to inconsistencies in coating thickness. Numerous workers are consistently exposed to potential hazards, including chemicals and dust, throughout the coating production line (Wojnar et al., 2020). To address these issues and usher in more effective coating techniques for steel components, there is a need to transition towards an automated process for coating H-shaped steel components. This shift involves employing an intelligent coating procedure, whereby robots are strategically guided to carry out the coating. Through meticulous planning of optimal coating paths, this approach not only mitigates direct operator exposure to the coating environment but also ensures uniform coating thickness across the surface of the component (Zgútová and Pitonák, 2021). By implementing automated coating, the problems associated with the manual method can be ameliorated, promoting safer working conditions and enhancing the overall quality of the coating process.

The crux of accomplishing coating tasks with intelligent coating lies in the synergy of pattern recognition and path planning (Chen et al., 2009). Pattern recognition serves as the ‘eyes’ of the coating robot, enabling the robotic arm to discern and gather information regarding the target, encompassing attributes such as location, shape, contour, size and other spatial particulars (Clément et al., 2018; Malu et al., 2018). Conversely, path planning functions as the cognitive core of the coating robot (Muzan et al., 2012), devising optimal trajectories for the robotic arm to yield maximum efficiency while executing tasks on the designated target. Hence, the central aim of this research was to establish streamlined coating paths for H-shaped steel components, effectively replacing manual labour. The significance of this study extends beyond resolving issues linked to manual coating methods, also ushering in heightened productivity.

Pattern recognition is intricately linked to the extraction of pertinent features or valuable details from images or objects, encompassing geometric shape, contours, texture, size and colour (Qi et al., 2023; Zhang et al., 2020, 2023a, 2023b). Establishing pattern recognition within an automated coating system is of paramount importance as it serves as the equivalent of the system’s visual capability to discern objects (Nguyen et al., 2020). Through pattern recognition, solid steel components can be comprehended, assessed and quantified on a computer using images and pre-existing models (Pan et al., 2023). This process subsequently generates and provides the geometric and coordinate data needed for the subsequent path planning (Gasparetto et al., 2010; Jingzhong et al., 2018).

With continuous advances in science and technology, the realms of manufacturing and processing are progressing toward mechanisation and automation, leading to a proliferation of practical applications for pattern recognition (Huang et al., 2016; Soualhi et al., 2020). Pattern recognition has been proposed as a means to capture scattering parameters for rough surfaces (Geisler and Kolb, 2018). For instance, Chen et al. (2019a) employed a sensor fence to recognise distinctive information from various parts of objects, displaying the recognised object image information on a computer screen. Statistical pattern recognition was harnessed by Balsamo et al. (2014) for structural health detection through damage identification algorithms. Qiao et al. (2012) used statistical pattern recognition along with a limited set of training data to detect structural damage. In the realm of continuous casting systems within the steel industry, Kempf and Adamy (2004) integrated pattern recognition using recursive fuzzy systems to serve as a monitoring mechanism. Lu et al. (2018) used existing measurements and pattern recognition to predict stress in distributed structural health monitoring. Ma et al. (2018) employed ground-penetrating radar for the automatic detection of corrosion in bridge steel bars. Pathirage et al. (2018) proposed structural damage identification through autoencoder neural networks and deep learning. In other work, artificial intelligence was employed for the assessment of steel bridge surface coating using a neuro-fuzzy recognition approach, a visual system was used to identify steel component attributes and the recognised images were used to optimise the best spray path plan (Bai, 2007; Chen and Chang, 2003).

The realisation of automated coating operations hinges on two fundamental factors: pattern recognition of the coated object and the generation of an optimal coating path. Within the framework of an automated coating system, the ability to devise an efficient coating trajectory significantly enhances the efficacy and precision of coating steel structures. Investigating optimised path planning, particularly in surface manufacturing or spray treatment, not only optimises the coating time but also ensures the uniformity of surface materials or coatings on objects (Armingol et al., 2003; Diao et al., 2009). The process of achieving efficient path generation broadly involves several steps, starting with acquiring dimensional data about the object, establishing a model of the object’s surface through pattern recognition, then devising an effective path considering coating conditions and ultimately computing the optimal path (Andulkar and Chiddarwar, 2015). Attaining the best path during the coating process necessitates an in-depth understanding of the execution procedure, along with an awareness of real-world constraints in the path planning stage (Chen et al., 2005a). In instances of spray coating using spray guns, a thorough grasp of the spray paint’s scope and geometry is pivotal, particularly when addressing issues of coating overlap along the path (Chen and Zhao, 2013; Chen et al., 2019b; Guan and Chen, 2019; Tang et al., 2014). Factors such as spraying parameters (e.g. air pressure, fluid pressure, spraying angle, spraying distance and movement speed) and the movement speed and acceleration limits of the robot arm joint impact the distribution of coating thickness, necessitating their consideration in path planning (Suh et al., 1991). The start and end points of the path play a critical role: they directly influence the path’s type and length, thereby impacting the calculation methodology of the path planning algorithm (Chen et al., 2005b).

It is imperative to consider the spacing between coating paths and the extent of variation in coating thickness or spray intensity distribution as part of the path planning process. These parameters significantly influence the consistency of the coating thickness distribution (Bruzl and Usmanov, 2016; Chen et al., 2017). In the context of coating methodology, conventional path design entails executing linear spraying followed by a transition to the subsequent linear trajectory. Thus, meticulous consideration must be given to the instructions for spraying and the choice of positions during this turning action within the path planning phase. Such deliberations hold direct sway over the uniformity of coating at the turning juncture (Duncan et al., 2005). Moreover, the orientation of the spraying movement – whether horizontal or vertical – directly dictates the length of linear spraying, the repetition count and the frequency of turns. This aspect must also be encompassed in the path planning due to its constraining influence (Trigatti et al., 2018).

Coating path planning involved the use of a shaft joint robot arm. Python software was employed for path programming, with the Matplotlib package facilitating the simulation display of the coating trajectory. The analysis and outcome verification phase considered coating parameters from Huiyang Machinery Co., Ltd, a steel component coating treatment factory in Taiwan. This information encompassed coating times, enabling a comparison between traditional and smart coating approaches. Initiating path planning entailed establishing the foundational structure of the path. Initially, an H-shaped steel component was subdivided into equidistant segments based on its overall length. Subsequently, the core movement mode of the robot arm – horizontal or vertical – was determined by specifying the length of a single linear motion and the number of cycles within the path. With the rudimentary path pattern in place, further path planning was undertaken by considering coating parameters, including the number of spray guns, spray time, spray range, coating thickness and path continuity. Following completion of efficient path planning, a simulation showcased the execution of the coating path on the H-shaped steel component’s surface. The process is summarised in Figure 1.

Considering the practicality of H-shaped steel components, instances where lengths exceed 10 m may pose challenges due to the limited reach of the robot arm. As a result, comprehensive coverage of the entire steel component might not be achievable. Therefore, when formulating path plans based on dimensional parameters, a systematic division of H-shaped steel components into uniform segments was adopted in accordance with their total length. During this segmentation process, given the uniformity of conditions and lengths across each segment, the path generated based on the prototype segment serves as a blueprint. This prototype path is systematically applied to other segments, ensuring consistent coating path construction across the entire structure. For the core motion strategy of the robot arm, an S-shaped trajectory was employed as the chosen spraying motion approach. Regardless of whether the motion is horizontal or vertical, the guiding principle is to maximise the length of each linear motion (individual pass) while minimising the number of cycles within the path (turns). By emphasising these two key aspects to regulate the number of turns in the path, the need for frequent halts caused by turns is minimised. The fundamental motion mode of the robot arm is illustrated in Figure 2.

Sequencing of the spraying process is determined by the number of spray guns employed. Two configurations were considered here – using a single spray gun and the use of three spray guns. Given that one side-of the H-shaped steel component is intended for integration with concrete, this side is typically left unpainted to enhance concrete adhesion and surface roughness. In the case of the single spray gun method, the left-hand section of the H-shaped steel component is coated first, followed by the upper/lower areas and then the right-hand section (Figure 3(a)). The specific sequence depends on the robot arm’s positioning. The three spray gun technique involves placing one spray gun in each of the three directions, allowing simultaneous coating of the left, upper/lower and right areas of the H-shaped steel component (Figure 3(b)).

The extent of spraying coverage and the thickness are established by configuring the spraying range and the velocity of the path. Users have the flexibility to fine-tune the path’s range and speed based on the specific capabilities of the spray gun and the desired thickness criteria.

An H-shaped steel component comprises three interconnected surfaces on its left and right sides. To ensure a seamless coating path, these surfaces are unfolded and flattened into an extended rectangular plane, as illustrated in Figure 4. This approach eliminates the necessity of repositioning the starting point for the subsequent surface after coating any of the three surfaces, streamlining the process.

The formulation of the coating path primarily entails representing the planned trajectory’s movement on an actual H-shaped steel component using simulation. During this phase of work, Tekla Structures (a comprehensive three-dimensional (3D) solid structure modelling software) was employed. This software not only automatically designs and generates installation drawings, component drawings and parts drawings, but also exports 3D object model files in *.STL format. This contour model is then transformed into a 3D representation using line segments. This transformation is achieved by extracting size parameters from the STL model files. Ultimately, the optimal path simulation was depicted on the contour model using the Matplotlib package.

The process of acquiring dimensional parameters for H-shaped steel components involved six distinct steps.

• ▪

  Step 1. Introduce the STL model files of H-shaped steel components into the algorithm for comprehensive analysis.

• ▪

  Step 2. Determine the plane generated by the plane equation using the three points that form the triangle within the STL model file. Subsequently, extract the normal vector of the plane. This normal vector serves as a reference point for subsequent elimination and categorisation.

• ▪

  Step 3. Remove duplicate planes that share the same plane equation on identical surfaces. During this step, only planes whose normal vectors align with the x, y or z axis in the spatial coordinates are retained.

• ▪

  Step 4. Organise the remaining planes into classes based on the x, y and z axes. Explore all conceivable combinations of length, width and height by altering the order of the x, y and z axes. Then, compare these combinations with the actual specifications of H-beams to identify the most suitable combination of length, width and height.

• ▪

  Step 5. Identify all feasible size combinations for H-shaped steel components. After establishing the location of each plane, leverage the classification of the three axes and consolidate length, width and height to compute the distance between two planes. This facilitates the derivation of all potential size combinations. Throughout this process, symmetry between the planes is considered and integrated into the calculation.

• ▪

  Step 6. Compare potential sizes against the precise specifications of H-shaped steel components to ascertain the most congruent size parameter outcomes.

Figure 5 shows the outcome of the size parameter analysis, revealing a length of 350 mm, height of 350 mm, width of 175 mm, web thickness of 7 mm, flange thickness of 11 mm and a joint radius measuring 13 mm. This analysis yielded calculated and scrutinised size parameter data, each parameter aligned with an appropriate representative label. In addition to the conventional descriptors of length, height and width, the representative names include cThick to signify web thickness, tbThick to denote flange thickness and radio to indicate the radius at the intersection of the web and flange.

With the primary path type determined, the Matplotlib package within Python software was used to generate a contour model based on the H-shaped steel component’s size parameters. This contour model served as the foundation for visualising and simulating the path’s comprehensive execution within a 3D space (see Figure 6). During this phase, the size parameter assumes a dual role – it not only contributes to contour model creation but also acts as a pivotal control factor governing the coating path’s movement. Using the size parameter, the painted surface’s boundaries are defined, facilitating informed adjustments and turns in the path. Furthermore, construction of the contour model enables the entire coating path to be depicted in spatial coordinates. Consequently, within this simulation, the alteration of coordinate positions visually illustrates the progression of the coating path.

The simulation outcome shown in Figure 6 illustrates the execution of the coating path using a single spray gun on the contour model. Path initiation was from the lower inner plane on the left, proceeding to the inner middle and upper planes on the left, then onto the inner upper plane on the right and finally maintaining a continuous trajectory towards the middle inner and lower planes on the right. Concerning the coating path’s execution with three spray guns, all three directions commence simultaneously. However, the sequence of executing the left and right internal planes is identical to the subsequent middle and upper plane execution. The simulation portrays the path as a line, with the rate of movement indicated by the spacing between lines (smaller gaps signify slower movement). The extent of coverage is represented by the length of the line.

Through rigorous testing of the coating path across various sizes of H-shaped steel components, it was established that successful acquisition of size parameters and contour models enabled the automatic planning of efficient coating paths. These paths were determined based on the obtained size parameters and were subsequently simulated using the contour model. An illustration of the coating path outcome for an H-shaped steel component with specifications of 300 × 150 × 6.5 × 9 mm is shown in Figure 7.

Upon finalising the design and establishment of the streamlined coating path for H-shaped steel components, the outcomes were subjected to verification. This verification process encompassed two key facets. The initial aspect entailed confirming the precision and adaptability of the efficient coating path across various dimensions of H-shaped steel components. The subsequent aspect involved a comparative analysis between the efficient path method and the conventional manual approach. The evaluation was conducted in terms of coating times to gauge the efficacy of the efficient path in optimising the coating process of H-shaped steel components.

To verify the accuracy and suitability of the efficient coating path, STL model files of various-sized H-shaped steel components were inputted to extract dimensional parameters. Using size parameters, the efficient coating path for each H-shaped steel component was then meticulously planned. This entailed constructing a contour model for each steel component, followed by simulating the coating path using the Matplotlib toolkit. The outcomes of the simulation served to validate the feasibility of executing the coating path for each H-shaped steel component. Furthermore, the simulation assessed whether the movement patterns of the paths aligned with the intended plan. If the path simulation encountered an issue, an examination of the simulation display programme was conducted, with adjustments made to address any relevant concerns. Subsequent path display simulations were conducted following these adjustments. Discrepancies between the displayed path and the intended plan triggered an evaluation of potential path type errors, pinpointing where and why the deviation occurred. The corresponding programme was then adjusted accordingly, and the path planning and generation process was iteratively re-executed until a precise and operational coating path for each H-shaped steel component was achieved. Ultimately, were carried out to analyse and enhance any problems encountered during the recording process. A flowchart of the initial phase of the verification process is provided in Figure 8.

For the second phase of verification, the times of the efficient coating path and the manual coating path were compared. In this verification, actual coating experiments were used to derive essential coating parameters, including the coverage range of a single spray, the configuration of the spray pattern and the velocity of the spraying motion. The experimental parameters were as follows: a spray range of 25 cm, a striped spray pattern and a spraying motion speed of 180 cm/s. The verification focused on two predominant sizes of H-shaped steel components: 700 × 300 × 13 × 24 × 2500 mm and 900 × 300 × 14 × 28 × 2600 mm. Following the acquisition of coating parameters and size specifications, the efficient coating path was devised and executed and the manual coating path was concurrently recorded. For these two sizes of H-shaped steel components, it became evident that the planned path was notably shorter than the manual coating path. Furthermore, the planned path exhibited a lower number of turning points.

For the 700 × 300 × 13 × 24 × 2500 mm component, the manual coating path required approximately 4.5 one-way paths to completion, whereas the planned coating path requires only three, as shown in Figure 9. In other words, the planned path required 33% less one-way paths than the actual manual coating path. Similarly, for the 900 × 300 × 14 × 28 × 2600 mm component, the manual coating path necessitated approximately 5.5 one-way paths, while the planned coating path required only four (Figure 10) – a reduction of approximately 27%. This reduction underscores the potential for enhanced efficiency in the coating process, curbing repetition and ensuring a more even coating thickness. Through an exhaustive examination of 99 samples, the duration of the planned coating path was found to be, on average, 24.6% less than that of the manual coating path (Table 1). This outcome substantiates the efficacy of the intelligent coating path.

The focus of this work was on the meticulous planning and establishment of an efficient coating path for H-shaped steel components. Path planning was a multi-faceted process, involving determination of the foundational path type and the consideration of pertinent coating parameters. Subsequent to acquiring size parameters using an algorithm from the STL model, Python software’s Matplotlib package was harnessed to craft contour models and delineate paths using line segments. These components collectively facilitated the simulation and visualisation of the execution of a streamlined coating path on the surface of H-shaped steel components.

The first part of results verification pertained to the planning and construction of the efficient coating path for varying sizes of H-shaped steel components. This phase substantiated the accuracy and applicability of the generated coating path outcomes. By comparing designed paths with the conventional manual approach, the proposed coating path translated to a reduction of approximately 20–35% in the actual manual coating path distance and an average time saving of 24.6%. Using the findings of this work, time savings and enhanced product quality in the coating process of H-shaped steel components can be achieved. This research on intelligent coating paths can provide a foundational framework for potential artificial intelligence applications in the manufacturing sector. Future work could integrate paint applicators for the protection of metal structures, focusing on evaluating productivity and material consumption. A deeper discussion on these topics could benefit practitioners by enhancing the efficiency of coatings production.

The authors extend their gratitude for the support for this research provided by the Taiwan Ministry of Science and Technology (MOST)/National Science and Technology Council (NSTC) under grant numbers MOST-108-2221-E-008-002-MY3, MOST-109-2622-E-008-018-CC2, MOST-110-2622-E-008-018-CC2, MOST-110-2221-E-008-052-MY3, NSTC-111-2622-E-008-017 and NSTC-111-2221-E-008-027-MY3. Any opinions, findings, conclusions and recommendations presented in this article belong solely to the authors and do not necessarily reflect the perspectives of MOST/NSTC.