The design of HDIRs is a highly specialized and technology-intensive field. It involves multiple disciplines, such as mechanical engineering, electronic engineering, control theory, and artificial intelligence [12]. There are different aspects of the design of HDIRs, which can be developed from two aspects: performance structure and morphology. In research on the performance structure of HDIRs, many scholars are committed to improving the working efficiency and stability of robots by enhancing components and conducting dynamic analysis [13]. For example, Chen et al. [14] addressed the issues of large driving torque and high energy consumption in HDIRs by modifying the transmission mechanism design and eliminating the balancing cylinder. They also combined kinematics and dynamics analysis to verify the effectiveness of the structural optimization, significantly reducing redundant energy consumption. Sun et al. [15] focused on the conceptual design of HDIRs and the improvement of overall performance, striving to improve the overall working efficiency of robots. In addition, Yin [16] significantly improved the handling efficiency of stacking robots through a comprehensive research method.

The structural design of HDIRs is a key factor in ensuring that the robot can withstand and operate under high loads. Current research on the body of HDIRs mainly focuses on the optimization of structural performance analysis. Chen et al. conducted research on the body structure design of HDIRs for specific loading, handling, and stacking scenarios. For example, a multi-specification cargo loading robot with automatic detection was designed, and the ANSYS Workbench platform was used to ensure that its telescopic mechanism met the working requirements [17]. Wei et al. used the Denavit-Hartenberg method and MATLAB, combined with ADAMS and ANSYS Workbench, to conduct kinematic and transient dynamic simulations of heavy-duty palletizing robots. Wu et al. [18] designed a material transfer robot that meets specific requirements and conducted structural and dynamic characteristic analysis[19]. Ko et al. [20] introduced the design process for heavy-duty handling robots. In a related study, Chung et al. [21] discussed the application of a specific analysis technology. Shang et al. [22] proposed a new drive form and established a kinematic model of a heavy-duty mobile robot. Miao et al. [23] completed a series of research from structural design to dynamic simulation for heavy-duty handling robots on production lines. These studies have achieved remarkable advancements in enhancing the structural performance of HDIRs, providing a strong guarantee for the stable operation of robots in complex industrial environments. However, they focus more on structural performance, while relatively less consideration is given to morphological design.

The morphological design of HDIRs has a pivotal impact on the human-machine interaction experience and the product’s market competitiveness. KE is a discipline that integrates users’ emotions and aesthetic preferences into product design [24]. The morphological design of HDIRs directly affects the usability and safety of human-computer interaction and plays a key role in brand awareness and market competitiveness [25]. To bridge the gap between technological rationality and user sensibility, it is urgent to introduce a methodology that can quantify and translate emotions and aesthetic needs into design elements [26]. As an interdisciplinary field, KE systematically translates user emotions and preferences into design decisions by integrating perception evaluation and styling. The purpose of KE is to integrate a product’s functionality with its aesthetics, thereby enhancing its market value and user experience. While KE has been applied in robotics to areas such as service robots [28], social robots [29], and human-robot collaboration [30, 31], its application to the specific field of HDIRs remains a largely underexplored area of research.

In existing industrial robot research, Zhou et al. [32] improved the efficiency of industrial robot morphological design in meeting users’ sensory needs by constructing a sensory image space. Chen et al. [33] introduced the KE theory and used the semantic difference method and the KAWAJI method to extract design elements from the three levels of sensation, perception, and association. Forming a semantic space provides a reference method for the design of industrial robot modeling. Based on KE, the design process and evaluation of industrial robots were further optimized [34]. Wang et al. [35] used qualitative and quantitative methods to explore the relationship between perceptual imagery and robot modeling characteristics. Taking a 7-degree-of-freedom industrial robot as the object, the design perception experience was enhanced through questionnaire surveys and eye-tracking technology. Taking a glass substrate transfer robot as an example, a modular design method was proposed based on KE [36,37]. Xiao et al. not only conducted an in-depth study on the morphological design of a dual-arm industrial robot, but also proposed an intelligent perception design method based on virtual reality and physiological signals. They also integrated Chinese traditional culture into product design, providing a new perspective for the innovation of modern mechanical product design [38–41]. Cheng et al. innovatively proposed the concept of an industrial robot perception intention space construction model based on the demands of users and experts. It promoted the in-depth participation of users and experts in all stages of the research [42]. KE is difficult to apply in the design field because it needs to process complex data and cognitive processes, which makes coding and measurement difficult [43]. It is not easy to integrate user perception needs into kinematic design. Benaissa et al. [44] found that user evaluation is affected by multiple factors. Integrating complex and ambiguous user needs into design is a challenge for designers. Previous studies on the morphological design of HDIRs have primarily focused on engineering and have rarely considered optimizing their form through KE. This results in the morphological design of robots often ignoring people’s subjective feelings and aesthetic needs while meeting industrial needs. It is challenging to stand out in the market competition, which is a significant shortcoming in current research.

In the field of morphological industrial robot design, the application of KE has gradually received attention. Through KE, the user’s emotional resonance and aesthetic value of the robot’s appearance can be fully considered, allowing for the design of products that better meet user expectations. Professor Nagamachi pointed out that with the correct application of KE, the quality and performance of the product can be ensured to reach a certain level, and emphasized that KE can guarantee a 100% product success rate [45].

AIGC technology refers to a new type of content production method that utilizes artificial intelligence technologies, such as deep learning, to automatically generate multimodal content, encompassing text, images, audio, and video through the learning of large-scale data [46]. AIGC technology has demonstrated significant potential in the field of industrial design. It is developing rapidly and is gradually becoming an important auxiliary tool for designers [47]. Currently, the representative software for AIGC-based image generation in the design field includes Stable Diffusion and Midjourney[48]. Stable Diffusion can quickly generate multiple high-quality design solutions based on the designer’s simple description or sketch, greatly improving design efficiency. At the same time, it can generate realistic product renderings in a short time, which helps designers intuitively display the design effect, facilitating communication with customers and solution optimization [49]. Midjourney can transform a sentence into a visually striking concept image at the lowest cost and the fastest speed, helping designers broaden their thinking, impress clients, and push creativity forward [50]. For example, Lu et al. developed a car design thinking model driven by AIGC technology. They thoroughly explored the significant role of this technology in the morphological design process of car products and the innovation and reformulation of the thinking model. Moreover, they verified the feasibility of the model through practical cases, effectively promoting the deep integration of artificial intelligence and the design field [51]. On this basis, Li et al. started from the perspective of manufacturing productivity transformation and, in combination with the actual situation of manufacturing in Guangdong Province of China, conducted a detailed analysis of the application potential of AIGC technology in product design [52]. Peng et al. [53] used the semantic analysis and metaphor expression techniques in the process of human-computer dialogue, and through dialogues with AIGC technology, utilizes the text-to-image technology to optimize the design process of high-end audio equipment.

In the field of industrial robots, AIGC technology can assist designers in training large models through text prompts or design plan manuscripts, generating a variety of design samples that provide designers with new sources of inspiration and multiple alternative design plans [54]. However, the application of AIGC technology in industrial product design is not without challenges. While AIGC technology faces challenges such as data set limitations and high training costs [55], it remains a tool with immense potential. Currently, its application to the morphological design of HDIRs is a largely unexplored area of research. If leveraged effectively, AIGC is poised to bring significant breakthroughs to this field of industrial design.

Based on the current research, the suggestions for the morphological design of industrial robots are largely general and lack specificity. In particular, there is an obvious research gap in the morphological design of HDIRs. Although some studies have begun to examine product design from the perspective of user emotions, they often ignore the importance of AIGC technology and corporate product genes in product design. In response to the above problems, this study took a six-axis 210kg welding HDIR as an example to guide the morphological optimization design.

The ray-firing method was initially limited to morphological transitions between two unexpanded curve segments. This study extended this method to arbitrary closed curves and constructs an innovative framework for the morphological optimization of industrial robots. This study demonstrates that the Ray-firing method is a powerful tool for innovative design, capable of retaining a brand's original visual characteristics to ensure brand consistency and preserve the value of brand assets in new industrial robot designs [51, 69].

Step 1: Use continuous cubic Bezier curves to draw two morphological curves (such as curve a and curve b). And obtain the feature point sets of the two curves, namely {a1, a2, a3, ..., a10} and {b1, b2, b3, ..., b12}. As shown in Fig. 1(a), curve a has 10 feature points and curve b has 12 feature points.

Step 2: Before blending the two curves, the two curves must have the same number of feature points. First, substitute the coordinate values of the feature points into formula (5) to calculate the centroid coordinates of curve a and curve b based on the centroid and overlap, as shown in Fig. 1(b). Second, curve b is the reference curve of curve a. And use the centroid as the origin to let the ray pass through the feature points of curve a. Finally, take the 12 intersection points (red points of curve a) and the ray as the new feature points of curve a, namely {a 1', a 2', a 3', ..., a 12'}.

Where and are the centroid coordinates, n is the number of feature points, x_i and y_i are the feature point coordinates.

Step 3: When the base curve and the reference curve have the same number of feature points, the new curve can be constructed using any formula in equations (1) to (4). For example, in the xy coordinate system, the feature point coordinates of the base curve, the new feature point coordinates of the reference curve, and the weight ratios of the five feature point sets (i.e., 3:1, 2:1, 1:1, 1:2, 1:3) are substituted into formula (1), i.e., plan 1, plan 2, plan 3, plan 4, and plan 5, and five curves can be constructed, as shown in Fig. 1(c).

DeepSeek is a high-performance large language model built with an advanced distributed training architecture [70]. In KE, DeepSeek can automatically generate high-fidelity affective emotional imagery adjectives based on users’ written prompts, providing scalable corpora for product emotion computing and significantly enhancing the consistency of intelligent assistants in dialogue and task processing. Recently, with its outstanding capabilities in large language models, DeepSeek has become one of the rapidly developing important applications [71]. DeepSeek’s continued progress is driven by continuous optimization, adaptation to various application scenarios, and frequent updates, such as the release of DeepSeek-v3 in October 2024.

Similar to text-to-image generation tools, prompt engineering is crucial to DeepSeek. On the Internet, many researchers and enthusiasts are actively engaged in prompt engineering practice [72]. They focus on discussing and sharing effective techniques for accurately using DeepSeek and exploring its functional limits. The key elements of DeepSeek prompt words include context, instructions, output indicators, and input data. For example, “I am conceiving an industrial robot of a Chinese enterprise brand (context). Please give the design concept (description), within 200 adjectives (output indicators), covering functions, appearance, and materials (input data).” At the same time, there are practical tips, such as “Please analyze as an expert (role)”, “Please imitate the style of (person or brand)”, and “Organize the following information in the specified format”.

In the field of natural language processing, DeepSeek-v3 and GPT-4.1 are two highly representative advanced language models [73]. GPT originated in the United States, and DeepSeek originated in China. The following is an in-depth analysis of the two from multiple key dimensions. The similarities and differences between the two are shown in Table 1. DeepSeek-v3 and GPT-4.1 are powerful multimodal language models, each with its own advantages [74]. In the field of industrial design, DeepSeek has great potential and is suitable for brainstorming, conceptual design, and competitive product analysis. DeepSeek can generate the prompts required for image generation artificial intelligence [75], so designers must master prompt engineering. Moreover, DeepSeek originated in China and excels at understanding descriptive vocabulary at various levels of abstraction, thanks to its excellent language processing capabilities. It aligns with the needs of Chinese companies, which makes it closely connected to KE theory and image generation tools such as Midjourney and Stable Diffusion.

KE is the systematic way of turning consumers’ feelings and emotional needs into concrete product features and functions [76]. Within this framework, phone interviews with business users play six key roles. First, they act like an emotional radar, quickly and cheaply cutting through company layers to capture what decision-makers really feel about product performance and brand image. Second, they serve as a semantic anchor, translating vague feelings into clear, countable words through semi-structured calls, giving the first raw data for later semantic-difference matrices. Third, they work as a dimension probe: by transcribing, coding, and counting words, we uncover hidden emotional angles and enlarge the space for principal-component analysis [77]. Fourth, they are a weight checker: short follow-up calls test how strongly each keyword resonates, correcting any mismatch between what people say and what the numbers say. Fifth, they act as an iteration trigger, feeding fresh market trends, competitor notes, and usage scenes into a live update loop that keeps generative models retraining. Sixth, they build a trust bridge, using frequent, low-friction contact to keep key users engaged so later prototypes and data flow run smoothly [78]. Phone interviews are the doorway to measuring emotions in KE and the live tuning knob that runs through every stage of user-centred innovation.

The standard five-step workflow is as follows. Step 1: Zero in on product pain points and core decision-makers, design a semi-structured, scenario-based guide, and run small-sample, in-depth interviews to catch spoken feelings in real time. Step 2: Transcribe the calls, build a multidimensional Kansei rating matrix using the semantic-differential method, code the raw words, count frequencies, and check reliability. Step 3: Use principal-component analysis to pull out the hidden kansei dimensions, pick statistically strong and highly distinctive keywords across three levels of emotional needs, and link them to quantifiable design variables like colour, material, and form [79]. Step 4: Break down key elements with morphological grammar, compute how much each variable shapes the intended image, and let large models quickly output prototypes that match the target feelings [80]. Step 5: Validate the prototypes’ emotional impact with follow-up calls, fold user feedback and market data into a dynamic update loop for continuous model improvement. In this way, phone interviews transform abstract feelings into actionable design parameters, establishing a method that is both theoretically sound and practically workable for user-centered industrial innovation [81]. At the same time, the resulting labels and weight maps also supply high-quality, scalable prompts for AIGC-driven design.

In 1987, Saaty proposed the analytic hierarchy process (AHP). It is a solution to decision problems involving multiple criteria [82]. Based on the fuzzy comprehensive evaluation method and the analytic hierarchy process, scholars proposed the fuzzy analytic hierarchy process, namely FAHP [83]. This method can effectively address the error problem caused by the traditional AHP method’s susceptibility to individual extreme values and respondent subjectivity [84]. Currently, FAHP has been applied to decision problems related to product design. The operation steps of FAHP are roughly the same as those of AHP [85]. The implementation steps are described as follows [86].

Step 1: Decompose the decision problem into the target layer, the guidance layer, and the indicator layer. Obtain user needs through questionnaire surveys and on-site interviews, filter and summarize them, and then build a model.

Step 2: Construct a fuzzy judgment matrix. Compare the indicators at the same level pairwise to determine the relative importance. The fuzzy judgment reference is shown in Table 2. Experts score the indicators according to the table and construct the fuzzy judgment matrix Q:

In the formula, r_ij represents the comparison between factor r_i and factor r_j, i, j = 1, 2, …, n.

Step 3: Calculate the indicator weight by the summation method and other methods. This method is simple and has high stability. The formula is as follows:

In the formula, W_i is the weight of indicator r_i, and n is the order of the matrix.

Step 4: Calculate the consistency ratio (CR). If CR ≤ 0.1, the judgment matrix is considered to have good consistency. The calculation formula is as follows:

CR calculation formula is as follows:

The random index (RI) values corresponding to matrices of different orders can be obtained by looking up Table 3.

Step 5: Weight ranking of demand indicators. Through comprehensive weight calculation, determine the weight ranking of each indicator relative to the target layer, and perform a consistency check again.

This study aimed to develop a morphological optimization design method for HDIRs based on KE, CBA, and AIGC technology and verify its effectiveness. This method optimizes the investigation, artificial generation, Curve blending, and visual output stages of the HDIRs design process. The complete implementation steps of the proposed design method are as follows.

Based on the design method and implementation procedures outlined in Section 3.2, the research framework is shown in Fig. 2. The implementation process of the generative artificial intelligence-based method for the morphology optimization design of HDIRs is described. The specific implementation process is as follows.

The study was conducted in accordance with the Declaration of Helsinki, and approved by the Institutional Review Board of the Beijing University of Technology (5 September 2024). First, we collected the emotional imagery adjectives of HDIRs morphology through traditional perceptual engineering enterprise user telephone interviews. Then, we used DeepSeek to generate the emotional imagery adjectives of HDIRs morphology.

As described in Section 3.2.2, the prompt method based on the Midjourney platform is described. The study uses the "/imagine" command to construct a prompt word template. The template consists of multiple structured elements, including dimensions such as subject description, morphological characteristics, design style, environmental settings, rendering parameters, and quality control. In response to the design requirements of the side morphology of HDIRs, the researchers once again used DeepSeek to construct multiple reference prompt words including core elements such as "210kg heavy-duty industrial robot", "Side profile of industrial robot", "Welding industrial robot", "Impressive/Streamlined/Stable/Powerful", "white clean background", "product design sketch", "complex details", "left view", "metallic yellow texture", and "dynamic" Multiple reference prompt words of core elements such as "state illumination", "3D", "C4D", "4K", "-- v 6.1", "-- ar 1:1", etc. Combined with the basic morphology of Fig. 7, the description skills of the general parameters of the product generated by uploading the image to generate text describe attributes. Through systematic parameter optimization experiments, the optimal prompt word structure was finally determined to be: "a/an [Emotional Imagery Adjectives] the 210kg heavy-duty welding industrial robot arm is placed in isolation in front of a white background. The left view effect drawing of the industrial product design shows a clean appearance with complex details highlighting the mechanical structure. The metallic yellow texture makes it suitable for product marketing materials or advertisements. Dynamic lighting, Turei Hattori style, 3D, 4K, C4D -- v 6.1 -- ar 1:1".

In the specific implementation process, this study selected four typical emotional imagery adjectives (Impressive/Streamlined/Stable/Powerful) from Section 4.1.3 as key variable parameters. By controlling the variable method, other prompt word elements were kept unchanged, and only the emotional imagery adjectives subitems were replaced for comparative experiments. At the operation level of the Midjourney platform, four initial morphological schemes can be generated after each emotional imagery adjective is input. The platform's builtin "variation" function can be further expanded to generate four derivative schemes. As shown in Fig. 4, users can implement iterative optimization of schemes through interface interaction. Click the function button in the lower right corner of the generated result to obtain new schemes of the same type. Select the "V" option with a specific number to generate similar schemes that evolve details while maintaining the core features.

Based on the above method, the study established a systematic industrial robot morphological image database. In the first round of generation, 8 initial schemes were generated for each of the four emotional imagery adjectives (a total of 32). On this basis, four variants were generated for each initial scheme, and finally a reference database containing 160 schemes (32 original schemes + 128 derivative schemes) was constructed. As shown in Fig. 6, taking the "streamline" image as an example, the eight basic solutions generated in the first round are numbered Int_f1 to Int_f8 (i.e., the first-generation morphological solutions). After selecting Int_f1 for variant generation, four optimized solutions with differences in morphological details, Int_f1_1 to Int_f1_4, can be obtained (i.e., the second-generation morphological solutions). This multi-level, iterative generation method effectively expands the diversity and selectivity of design solutions.

In order to clearly explain the proposed design method, this study uses a 210kg Chinese-made welding HDIR that needs to optimize its appearance as the basic model. As shown in Fig. 6, this model was selected as the basic form. Its forearm is bloated, the curves of the upper and lower arms do not match, and the overall uniformity is poor. The study aims to improve the overall consistency of its appearance through design improvements. The study selected the top two typical emotional imagery adjectives "Impressive" and "Streamlined" as the target image of the basic form. Further clarify the design goal: on the basis of retaining the core product gene characteristics of the basic form, create a new form that matches the specified target image.

Given that the target images set by the basic form are "Impressive" and "Streamlined", the visual features of the reference form should also show the characteristics of "Impressive" and "Streamlined". Based on this, this study once again invited 6 experts (2 males and 4 females) as participants. Through open discussions, typical forms that can represent "Impressive" and "Streamlined" were selected from the reference database. Specifically, the 6 participants were divided into two groups. Each group selected 4 images with both "Impressive" and "Streamlined" features from the first-generation database. Subsequently, based on the selection results of the two groups, all participants jointly selected representative images from the corresponding second-generation database. Finally, after collective deliberation, the participants unanimously selected Imp_f1_1 and Str_f1_4 (see Fig. 5) as the typical representative forms of the target image, which will be used as reference forms for subsequent designs.

In order to fuse the basic morphology with the reference morphology, we use three consecutive Bezier curves to draw the morphological element curves, covering the Main Body, Upper Arm, Lower Arm, Base, End Effector, and Joints. This study uses Rhino to draw the morphological element curves. In addition, in order to ensure the effectiveness and usability of the morphological element curves in morphological fusion, we normalize the overall width, as shown in Fig. 7.

Six experts were invited again to use the shape analysis method to deconstruct and analyze the appearance of the HDIRs. Its appearance was subdivided into six key design features: covering the main body, upper arm, lower arm, base, end effector and joint components. In the specific operation, the basic form, reference form 1 and reference form 2 were further divided. The five shape element curves under each form were extracted as shown in Fig. 8. After completing the decomposition operation, with the help of the curve analysis function of the Rhino software, the number of feature points contained in each shape element curve and its corresponding coordinate values were accurately obtained. Next, these coordinate values were substituted into formula (5) to calculate the centroid position of each shape element curve.

In order to achieve the fusion of the morphological element curves between the basic shape and the reference shape, the study adopted a curve blending algorithm combined with a ray emission method. Taking the blending of the basic shape and the reference shape 1 or 2 as an example, Fig. 9 reference form 1 shows the blending effect of curves a_0 and a’_0. The fusion steps are as follows: First, the position alignment is completed according to the centroid of curves a_0 and a’_0. Then, curve a_0 is selected as the reference curve and curve a’_0 is selected as the target curve. Starting from the centroid, 90 rays are emitted along the 90 feature points of curve a_0. Next, the 90 intersection points of curve a’_0 and the rays are captured and used as the new feature points of curve a’_0. Using formula (1), the feature point sets of the two curves are combined with different weights (3:1, 2:1, 1:1, 1:2, 1:3) to generate 5 new morphological curves, named a’_1 to a’_5 in sequence. The coordinates of the feature points of these 5 new curves are further substituted into formula (5) to calculate the centroid position of each curve. Although the newly generated curves take into account the morphological features of the basic shape and the reference form, some curves are not smooth. This may be because the new curves do not inherit enough feature points or lose some important feature points, which requires designers to repair and smooth them. Through such mixed operations, the study derived 25 new curves in each morphological element, providing morphological materials for subsequent design.

The morphological element curves generated by the basic shape and the reference morphology are combined into 10 new morphologies based on the centroid of the reference morphology, as shown in Fig. 10. Based on the centroids of the 6 morphological element curves (i.e., from C0 to C6) of reference morphologies 1 and 2, 5 groups of morphological element curves are combined into 10 plans, namely, plan_1- plan_10. Taking plan_1 as an example, it includes curves a’_1, b’_1, c’_1, d’_1, e’_1 and f’_1.

In order to present the design plan more quickly and realistically, Stable Diffusion and controlnet are used to control the generation of 10 plans' 2D line drawings into 3D renderings. As shown in Fig. 11, the renderings of the ten plans present realistic visual effects with a high degree of restoration. In addition, the image generation process needs to pay close attention to the effect, flexibly adjust the redrawing amplitude, control weight and other parameters, and continuously optimize the prompt words. Through repeated trials and iterations, the quality of the generated image is gradually improved, and finally a 3D image that meets the expected specific contour line is obtained.

Two evaluation methods were used for the evaluation of the 10 Plans (Fig. 11): expert evaluation based on FAHP (Evaluation 1) and consumer perception evaluation (Evaluation 2). In Evaluation 1, 6 experts were invited to evaluate the 10 morphological Plans generated by morphological fusion using FAHP. The evaluation criteria were "How impressive/streamlined is Plan_1 compared with Plan_2? ". The corresponding pairwise comparison matrices of Evaluation 1 are shown in Table 6. For the impressive Plan_1–5, ranked from the highest to the lowest in terms of the comprehensive weight, they are Plan_5, Plan_4, Plan_3, Plan_2, and Plan_1; for the streamlined Plan_6–10, ranked from the highest to the lowest in terms of the comprehensive weight, they are Plan_9, Plan_8, Plan_7, Plan_6, and Plan_10. The results of the first assessment show that the most ideal solutions are Plan_5 and Plan_9, and the C.R. values for both assessments are less than 0.1, indicating that the assessment results are acceptable.

In evaluation 2, 5 enterprise users who participated in the KE-based interview and 40 relevant colleagues were invited to conduct telephone interviews to evaluate the 10 so-lutions based on the standard of “To what extent does this plan represent your emotional feelings about the side profile of a 210kg impressive/streamlined heavy-duty welding industrial robot ? ” The score ranged from 0 to 1. A total of 43 valid data sets were collected, including 35 males and 8 females, aged between 28 and 55 years old. After statistical analysis, Cronbach's α coefficient exceeded 0.8, indicating that the results were acceptable. The results are listed in Table 7. The experts and enterprise users have the same ranking for the impressive Plan_1–5; and they also have the same ranking for the streamlined Plan_ 6–10. In addition, the Pearson correlation coefficient was used to test the correlation between the two groups of evaluation results. The analysis showed that the correlation coefficient exceeded 0.9 and the p value was less than 0.01, indicating that there was a significant and statistically robust correlation between the two evaluation results. In the two evaluations, the ranking of the schemes remained consistent and was proportional to the curve fusion weight value of the reference form. This means that the higher the blending weight value of the reference curve, the more the scheme can be consistent with the image target feeling of the reference form.

Integrating KE, CBA, and AIGC, this study conducted an in-depth optimization of the morphology of HDIRs, successfully preserving the visual identity inherent in existing product curves. Furthermore, the proposed morphological optimization design method offers designers a novel and streamlined workflow experience.

Moving beyond the limitations of traditional design approaches, which often rely heavily on designers' subjective judgment[87], this research employed a ray emission method to blend existing product curves with functional requirement curves, generating a variety of innovative design schemes. This methodology enables the simultaneous consideration of product curve genetics and the precise matching of functional demands, thereby producing morphological solutions that integrate brand identity with practicality. This innovative design approach effectively resolves the traditional conflict between heritage, requirements, and utility. It provides new perspectives and methodologies for the field of HDIRs design, promoting the comprehensive advancement of design philosophy.

By leveraging advanced AIGC technologies such as Deepseek-v3, Midjourney, and Stable Diffusion, significant efficiency improvements were achieved within the morphological design process of the 210kg welding HDIRs. Compared to traditional design processes, the design cycle can be shortened while the number of design iterations increases substantially[89], These powerful AI tools rapidly generate a large number of creative morphological design proposals for HDIRs[90]. This facilitates the exploration of a wider range of design possibilities within a shorter timeframe, establishing a more efficient paradigm for the morphological optimization of HDIRs.

The research integrates the KE-CBA fusion strategy and the ray shooting method into the morphological optimization design process of HDIRs and applies it to a 210kg welding HDIR. This method is simple and can be implemented in medium and small-sized industrial robot enterprises without overly complex procedures. Moreover, the industry association can incorporate this method into the design guidelines for HDIRs, promoting and publicizing it through case libraries and training materials, thereby reducing iteration costs and shortening the market launch cycle of domestic HDIRs models, providing a replicable design path for batch customization and brand differentiation.

This study has yielded preliminary findings while acknowledging several limitations that must be addressed in subsequent research. First, the current investigation was confined to 210kg welding HDIRs, characterized by specific working loads and motion trajectories. Consequently, the methodology’s generalizability remained unverified across diverse operational contexts. Future research should extend to applications with substantially different performance requirements, including palletizing and ultra-heavy-load (≥ 3 tons) scenarios, to systematically evaluate the framework's adaptability and robustness under varied working conditions. Second, the AIGC training dataset exhibited deficiencies in specialized industrial image modeling, lacking comprehensive representation of complex and diverse structural features inherent in real industrial environments. This shortcoming may introduce discrepancies between generated solutions and engineering practicalities, potentially compromising the precision and reliability of design outcomes. Consequently, developing a high-quality, specialized industrial image dataset represents a critical objective for future work. Last, the present research primarily focused on conceptual and schematic design phases, without adequate consideration of production constraints such as casting requirements. Furthermore, AIGC-generated structures frequently conflict with manufacturability imperatives. The crucial future direction involves deeply integrating AIGC structural optimization with mass production guidelines, establishing a design optimization workflow that incorporates manufacturing process constraints. This approach will ensure innovative designs achieve morphological excellence and industrial feasibility, thereby facilitating effective translation from theoretical design to engineering application.