Method for generating custom connectors using an ai generator

Abstract

The invention provides a Method for generating custom connectors using an Al generator comprising the steps of: - Providing some input information to a LLM wherein, the input in-formation comprises human natural language information; - Processing the input information with the LLM into a set of specification that is interpretable by a computer; - Aligning the set of specification towards physical requirements; - Using a generative design algorithm for generating an optimized 2d and / or 3d model set of data incorporating the set of specifications and the physical requirements; - Outputting a 2d and / or 3d set of data provided by the custom connector model.

Description

[0001] Method for generating custom connectors using an Al generator

[0002] Description:

[0003] The invention relates to ease a creation of custom connectors based on customer specifications.

[0004] The purpose is to offer shorter time-to-market and faster iterations based on a customer feedback.

[0005] Todays, process is based on initial designs by educated engineers that are refined and iterated using software simulation and / or testing physical prototypes. This procedure is very time-consuming and requires a great amount of manual work.

[0006] The present invention addresses the challenge of generating custom connectors automatically.

[0007] The invention solves this problem through a method according to claim 1 .

[0008] According to the invention, this is achieved by a method for generating custom connectors using an Al generator according to claim 1 , a product data file according to claim 10, a computer program product according to claim 11 a computer system according to claim 12 a method for producing a custom connector according to claim 13.

[0009] Advantageous embodiments can be taken from the sub-claims, for example. The content of the claims is made the content of the description by express inclusion.

[0010] The invention provides a method for generating custom connectors using an Al generator, comprising the steps of: Providing some input information to a Large Language Model (LLM) wherein, the input information comprises human natural language information;

[0011] Processing the input information with the LLM into a set of specification that is interpretable by a computer;

[0012] Aligning the set of specification towards physical requirements;

[0013] Using a generative design algorithm for generating an optimized 2d and / or 3d model set of data incorporating the set of specifications and the physical requirements;

[0014] Outputting a 2d and / or 3d set of data of the custom connector model.

[0015] In the context of the invention, the described method is a computer-implemented method designed to be executed by a computer.

[0016] The invention offers significant advantages in the field of custom connector design and manufacturing. By leveraging artificial intelligence and machine learning techniques, this method streamlines the process of creating bespoke connectors tailored to specific customer requirements. The use of a Large Language Model (LLM) to interpret human natural language input allows for a more intuitive and accessible interface, enabling customers to describe their needs without requiring extensive technical knowledge.

[0017] The automated processing of input information into computer-interpretable specifications significantly reduces the time and effort traditionally required for manual translation of customer requirements. This automation minimizes errors that can occur during manual interpretation and specification creation. The alignment of specifications with physical requirements ensures that the generated designs are not only theoretically sound but also practically feasible for manufacturing.

[0018] By employing a generative design algorithm, the invention explores a vast design space rapidly, considering numerous iterations and optimizations that would be impractical for human designers to evaluate manually. This results in innovative connector designs that are optimized for multiple criteria simultaneously, such as performance, cost, and manufacturability. The output of optimized 2D and / or 3D model data sets provides a direct link to manufacturing processes, potentially reducing the time from concept to production.

[0019] Furthermore, this method enhances the iterative design process, allowing for quick adjustments based on customer feedback or changing requirements. The speed and efficiency of the Al-driven approach enables faster prototyping cycles, accelerating the overall product development timeline. This acceleration in design and production provides a competitive advantage in the fast-paced electronics and connectivity markets. The invention also contributes to cost reduction in the custom connector development process by minimizing the need for extensive manual engineering hours and physical prototyping. By simulating and optimizing designs digitally before physical production, the method reduces material waste and associated costs. Additionally, the consistency and repeatability of the Al-driven process leads to higher quality standards and more reliable outcomes in custom connector design.

[0020] A prime example of a connector type that can be customized easily is the PushPull connector system by manufacturer Harting. It is known from the website: https: / / www.harting.eom / de-DE / s / splp-harting-pushpull and from the brochure "HARTING Device Connectivity", publication number MO / 2016-03-01 / 5.75 98 41 007 0201 Version 07 5, pages 66-137. This connector offers a range of easily handled housings for transmitting data, signals, and power across various applications. The modular concept includes standard RJ45 and USB interfaces as well as connector faces for signals, power transmission, and optical interfaces. Various metal or plastic materials can be used. The system provides connectors for signal, power, and optical interfaces in simple-to-use housings suitable for different applications. It is designed as an solution for Ethernet applications, featuring a robust housing.

[0021] The PushPull system includes cable assemblies, connector sets with crimp termination, and device connection technology. It offers solutions for demanding outdoor applications such as camera systems for monitoring process plants, telecommunications equipment, and traffic technology. These applications expose connectors to environmental factors like sun, wind, rain, and salt spray in coastal areas.

[0022] The PushPull connector system incorporates various connector types, including those for display interfaces and robotics applications.

[0023] Another interesting connector type for customization are cable to board connectors, especially those with angled pins shown on page 864 left column of the aforementioned brochure "HARTING Device Connectivity". For these connectors, the inventive method is particularly useful as it allows for an optimization of not only the pins of the plug but also its counterpart (socket holes) in the board.

[0024] It has been appreciated that a method is needed that overcomes one or more of these problems.

[0025] The input information can be a written submission from a chat bot or a speech to text input via a microphone or any other written or digital input. Large Language Models (LLMs) are types of artificial intelligence (Al) models designed to understand, generate, and sometimes translate human language. They are "large" in the sense that they consist of many parameters (which can range from hundreds of millions to hundreds of billions), enabling them to capture a vast amount of information about language patterns, grammar, context, and semantics.

[0026] Several current Large Language Models (LLMs) may be applied for generating custom connectors, including both open-source and proprietary versions. These models include the chatGPT family from OpenAI, BERT from Google, and LLaMA models from Meta Al. These models may be adapted and fine-tuned for specific tasks related to generating custom connectors, with the choice of model depending on factors such as the specific requirements of the task, computational resources available, and licensing considerations.

[0027] The LLM can be fine-tuned specifically for connector design tasks through a multi-stage process that enhances its domain expertise and technical accuracy. As used herein, the term "fine-tuning" may refer to a process of adapting a pre-trained language model to a specific task or domain by further training it on a smaller, task-specific dataset. Fine-tuning may involve adjusting the model's parameters to optimize performance on the target task while retaining the general language understanding acquired during pre-training. This process may modify the weights of some or all layers in the neural network, potentially resulting in improved accuracy, relevance, and specificity for the intended application. Fine-tuning may allow the model to learn domain-specific vocabulary, conventions, and reasoning patterns, which may enhance its ability to generate more accurate and contextually appropriate responses in the target domain.

[0028] This fine-tuning begins with collecting a comprehensive dataset of connector specifications, technical documentation, industry standards (such as IEC, UL, CSA, and MIL-STD), and historical design cases that represent successful custom connector implementations. The training corpus should include detailed connector datasheets, engineering drawings, material property databases, electrical performance characteristics, and documented design rationales from previous projects. Domain-specific pre-training can be implemented by exposing the model to connector engineering literature, technical manuals, and industryspecific terminology. This phase helps the LLM develop a specialized vocabulary and conceptual understanding of connector design principles, manufacturing constraints, and performance requirements. The model can then undergo supervised fine-tuning using paired examples of natural language requests and their corresponding technical specifications, where human experts provide the ground truth translations of customer requirements into precise engineering parameters.

[0029] Reinforcement learning from human feedback (RLHF) can further refine the model's capabilities by incorporating evaluations from experienced connector design engineers. In this approach, multiple specification outputs are generated for the same input, and engineers rank or score them based on technical accuracy, completeness, and feasibility. These rankings train a reward model that guides the LLM toward producing more technically sound specifications.

[0030] To enhance contextual understanding, the model can be trained with multi-modal inputs that include not only text descriptions but also images of similar connectors, circuit diagrams, or application environments. This helps the LLM better comprehend the physical and operational context in which the connector will function. The fine-tuning should also include adversarial training examples that present challenging or ambiguous requirements, teaching the model to identify when additional information is needed and to generate appropriate clarifying questions.

[0031] Continuous learning mechanisms can be implemented to allow the model to improve over time based on feedback from actual implementation outcomes. When a generated specification leads to a successful connector design, this positive outcome reinforces the model's approach. Conversely, when specifications require revision, these corrections are fed back into the training process. The fine-tuned LLM should also be integrated with a knowledge retrieval system, for example with internet access to fetch the latest standards definitions, material properties, and manufacturing capabilities to ensure that generated specifications remain current with industry developments and technological advancements.

[0032] The generated set of specifications can be reviewed by a human expert to ensure that they are technically sound and meet all the required standards and user needs. The LLMs might not fully grasp the engineering complexities or safety standards involved in designing such components.

[0033] Prompting patterns are structured approaches to formulating inputs for LLMs to elicit specific types of responses. For custom connector design, several prompting patterns can be employed:

[0034] Interview-style prompting is particularly effective for gathering comprehensive specifications. This pattern involves the LLM asking a series of questions to collect all necessary information before proceeding to the next step in the design process. Chain-of-thought prompting guides the LLM through a logical sequence of questions, where each subsequent question builds upon previous answers to create a complete specification profile. This ensures all interdependent parameters are properly considered. Tree-based prompting allows the LLM to branch its questioning based on previous responses. For example, if a user specifies an outdoor connector, the LLM would follow up with environmental resistance questions that wouldn't be relevant for indoor applications. Checklist prompting ensures the LLM systematically covers all required specifications by working through a predefined list of necessary parameters, asking about each one until the checklist is complete.

[0035] For an effective interview mode, the LLM can implement these patterns for example as follows:

[0036] Initial broad question: "What type of connector are you looking to design?"

[0037] Follow-up specific questions:

[0038] - "How many pins or contacts will this connector require?"

[0039] - "What voltage and current ratings are needed?"

[0040] - "What environmental conditions must this connector withstand?"

[0041] The LLM tracks which specifications have been provided and which are still missing, continuing to prompt until all required information is collected.

[0042] Validation prompting can be incorporated where the LLM repeats back the understood specifications and asks for confirmation: "I understand you need a waterproof circular connector with 8 pins rated for 24V DC. Is this correct, or would you like to modify any specifications?"

[0043] Priority-based prompting ensures critical parameters are addressed first before moving to optional features: "First, let's establish the essential electrical requirements before discussing physical dimensions and special features."

[0044] Contextual prompting allows the LLM to adapt its questions based on industry standards or typical use cases: "Since you mentioned aerospace applications, I'll need to ask about vibration resistance and temperature range specifications that meet industry standards."

[0045] Completion checking prompts help the LLM verify that all necessary information has been gathered: "I've collected information on electrical ratings, physical dimensions, and environmental requirements. Before proceeding, are there any special features or compliance standards we haven't discussed?"

[0046] The interview mode continues until the LLM determines it has gathered all specifications needed for the next design phase, at which point it can transition from information gathering to information processing and design generation.

[0047] By implementing these prompting patterns in an interview mode, the LLM creates a conversational interface that systematically collects all required specifications while remaining flexible to the user's level of technical expertise and specific requirements.

[0048] The interview process for generating custom connectors can also be conducted through voice interaction, allowing users to speak directly to the Al tool without typing. This voicebased interface may enhance user convenience and accessibility, particularly in situations where typing is impractical or for users who prefer verbal communication. In a voice-enabled system, the Al tool may incorporate speech recognition technology to convert spoken words into text. This speech-to-text conversion may allow the Large Language Model (LLM) to process verbal inputs in a manner similar to written text. The system may utilize natural language processing (NLP) techniques to interpret the user's spoken instructions and queries.

[0049] For responding to the user, the Al tool may employ text-to-speech (TTS) technology. This may enable the system to convert its text-based responses into audible speech, creating a fully interactive voice dialogue. The TTS system may be designed to produce naturalsounding speech with appropriate intonation and pacing to enhance comprehension. The voice interface may support the same prompting patterns and interview techniques previously described. For example, the Al tool may verbally ask the initial broad question: "What type of connector are you looking to design?" The user can then respond verbally, and the conversation may continue with follow-up questions and clarifications. Multilingual support to accommodate users who prefer to speak in languages other than English can also be applied. The voice-enabled Al tool may also provide visual feedback or display key information on a screen, creating a multimodal interaction experience. This may help users verify that their spoken inputs have been correctly interpreted and allow them to review the gathered specifications visually.

[0050] In cases where precise technical terms or complex specifications are involved, the system may offer a hybrid approach. Users may speak their general requirements and then refine or confirm specific details through on-screen prompts or by spelling out critical information. By offering both text-based and voice-based interaction options, the Al generator for custom connectors may accommodate a wider range of user preferences and usage scenarios, potentially improving accessibility and user experience in the connector design process.

[0051] An example of the set of specifications that can be outputted from the LLM based on user input can be:

[0052] Connector Type: Circular DIN

[0053] Number of Pins: 7

[0054] Rated Voltage: 250V

[0055] Rated Current: 5A per pin

[0056] Contact Resistance: <5 mOhm

[0057] Insulation Resistance: s100 MOhm at 500V DC

[0058] Operating Temperature Range: -40°C to 85°C Ingress Protection Rating: IP67 (dust tight and water-resistant)

[0059] Shell Material: Nickel-plated brass

[0060] Contact Material: Gold-plated copper alloy

[0061] Termination Type: Solder cup

[0062] Compliance: RoHS Directive, CE

[0063] The information generated by the LLM may be organized and stored in a structured data format resembling a table or matrix. This format may utilize pre-defined technical terms as categories, ensuring that the information is presented in a standardized manner. By employing such a structure, the system may facilitate easier parsing, comparison, and analysis of connector specifications across different designs. The standardized format may include columns for each technical parameter, such as connector type, number of pins, voltage ratings, current ratings, resistance values, temperature ranges, protection ratings, materials, and compliance standards. Each row in this table-like structure may represent a unique connector specification set based on the user input, allowing for efficient storage and retrieval of multiple design outputs. This approach to data organization enhances the system's ability to quickly access and compare different connector designs, streamlining the decision-making process for engineers and designers. Furthermore, the standardized format improves interoperability with other design tools and databases, enabling seamless integration into broader engineering workflows and systems.

[0064] In another embodiment the LLM is configured to validate the input information according to at least one of the following parameters: value range; threshold value;

[0065] - sign; contradictions.

[0066] The validation process implemented by the LLM serves as a critical quality control mechanism to ensure technical feasibility and coherence of the custom connector specifications. When validating value ranges, the LLM examines whether specified parameters such as voltage ratings, current capacities, or dimensional measurements fall within physically possible and industry-standard boundaries. Threshold value validation ensures that minimum requirements for performance characteristics like insulation resistance or contact durability are met or exceeded. Sign validation prevents logical errors by confirming that directional values are correctly specified (e.g., positive values for resistance measurements). Contradiction detection represents perhaps the most sophisticated aspect of validation, wherein the LLM identifies and flags mutually exclusive requirements, such as specifications for both high current capacity and miniature form factor that would create thermal management issues, or material selections incompatible with stated environmental resistance requirements. This comprehensive validation framework significantly reduces design iterations by identifying potential issues before the physical modeling phase begins.

[0067] The LLM could be configured and / or trained to correct incorrect and / or incomplete input information. The LLM comprises and / is trained with a set of training data comprising a number of predesigned plug connectors. The LLM can be configured to access a look-up table information and / or other external information.

[0068] The LLM can be further configured to refine the set of specifications by inputting a new or amended prompt.

[0069] In an embodiment the Al generator is configured to be executed on promise or to out-source tasks to a cloud if the capacity utilization of the on prmise instance exceeds a certain threshold, wherein the threshold takes into account at least one of the following metrics: percentage of total available CPU untilization, GPU memory consumption, RAM usage, number of floating-point operations per second, total duration of running a task. The capacity utilization threshold that triggers cloud outsourcing can be defined along multiple dimensions. Primarily, it may be expressed as a percentage of total available computational resources, typically ranging from 75% to 90% of CPU utilization, GPU memory consumption, or RAM usage. For example, when the local system detects that CPU utilization has sustained above 85% for a predetermined period (e.g., 5 minutes), it may initiate the cloud outsourcing protocol. The threshold can also be defined in absolute terms, such as when processing requirements exceed a specific number of floating-point operations per second (FLOPS), typically in the range of teraFLOPS for complex generative design tasks. Time-based thresholds may also be implemented, where tasks estimated to require more than a certain duration (e.g., 2 hours) on local hardware are automatically routed to cloud resources to maintain system responsiveness for other users or tasks.

[0070] When the threshold is exceeded, the system employs a resource allocation algorithm to determine which specific components of the workflow should be outsourced. The LLM processing might remain on premise for data security reasons, while computationally intensive simulations and generative design iterations are distributed across cloud resources. This hybrid approach optimizes both performance and cost-effectiveness, as cloud resources typically incur usage-based fees. The system maintains continuous monitoring of both local and cloud resource utilization, dynamically adjusting the distribution of computational tasks based on real-time performance metrics, priority levels of different design projects, and cost considerations. Additionally, the system implements data synchronization protocols to ensure that all design information remains consistent across the distributed computing environment, with appropriate encryption and security measures applied to data in transit between onpremise and cloud systems.

[0071] The training data comprising predesigned plug connectors serves as a foundational knowledge base for the LLM, enabling it to recognize patterns, specifications, and design constraints common in connector engineering. This dataset typically includes comprehensive technical documentation of existing connector designs, including dimensional specifications, electrical parameters, mechanical properties, material compositions, and performance characteristics across various operating conditions. When processing user input, the LLM leverages this preexisting knowledge to identify similarities between the requested custom connector and previously designed connectors, allowing it to suggest appropriate specifications even when user input is ambiguous or incomplete. For instance, if a user requests a "waterproof connector for marine applications" without specifying an ingress protection rating, the LLM can draw upon its training data to recognize that marine environments typically require at least IP67 or IP68 protection and suggest this specification accordingly.

[0072] The look-up table integration provides a structured reference mechanism that complements the LLM's neural network capabilities with precise, deterministic data retrieval. These tables are meticulously organized databases containing standardized specifications for connector components, industry-standard parameters, material properties, and regulatory compliance requirements. When the LLM processes user requirements, it can query these look-up tables to retrieve exact values for specific parameters, ensuring technical accuracy in the generated specifications. For example, when a user specifies a connector for a particular application, the LLM might access look-up tables containing standard pin configurations, voltage ratings, and current capacities appropriate for that application domain. The look-up tables also contain relational data that helps the LLM understand interdependencies between different specifications, such as how increasing current capacity might necessitate larger conductor cross-sections or how certain material choices might impact temperature resistance.

[0073] The structured data table obtained from the user interview may be systematically compared to the look-up table to validate and refine the specifications. This comparison process may involve matching each parameter from the user-provided data against corresponding entries in the look-up table. The system may check if the user-specified values fall within acceptable ranges defined in the look-up table, or if they align with standardized options for that particular connector type or application. In cases where the user-provided data differs from standard values, the system may flag these discrepancies for further review or automatically suggest the closest matching standard specification. This comparison step may also help identify any missing critical parameters that the user may have overlooked, prompting the system to request additional information or propose default values based on similar connector designs in the look-up table. By cross-referencing the user input with standardized data, the system may ensure that the final connector specifications are both technically feasible and compliant with industry standards.

[0074] The synergistic interaction between the pretrained connector knowledge and look-up table functionality creates a robust system for specification refinement. When the LLM encounters incomplete information, it first attempts to infer missing specifications based on its pretrained understanding of connector design principles and typical configurations. It then validates these inferences against the look-up tables to ensure they conform to industry standards and physical feasibility constraints. This validation process involves cross-referencing multiple tables to check for consistency across interdependent parameters. For instance, if the LLM infers a certain contact material based on the application environment, it will check material compatibility tables to ensure this material choice is appropriate for the voltage, current, and temperature conditions specified elsewhere in the requirements. This multi-layered approach to specification completion ensures that even when user input is minimal, the resulting connector design will meet all necessary functional requirements and comply with relevant industry standards.

[0075] The LLM can be further configured to refine the set of specifications by inputting a new or amended prompt. The computational requirements for executing the Al generator on premise can be substantial, particularly when processing complex connector designs with numerous parameters and constraints. The system typically requires high-performance computing hardware with multi-core processors, substantial RAM capacity, and dedicated graphics processing units (GPUs) to efficiently handle the computational demands of both the Large Language Model and the generative design algorithm. For instance, the LLM component may require 16-32GB of RAM for optimal performance, while the generative design algorithm, especially when performing finite element analysis or other simulation tasks, may require additional dedicated computational resources. Storage requirements are also significant, as the system must maintain databases of connector specifications, material properties, and historical design data.

[0076] In another embodiment the LLM is configured to output the processed information from the human natural language information comprising the set of specification to a user. This output can be presented during the interview mode, where the LLM displays the interpreted specifications on screen as they are being collected through the text-based or voice-based interaction. For text-based interviews, the specifications may appear in a structured format alongside the conversation, allowing users to visually confirm the accuracy of the interpreted requirements in real-time. In voice-based interactions, the system can display the recognized specifications on screen while simultaneously providing verbal confirmation through text-to- speech technology. This multimodal presentation enables users to review and verify the processed specifications during the interview process, facilitating immediate corrections or adjustments before proceeding to the design phase.

[0077] In an embodiment the physical requirement is aligned by a look-up table and / or algorithm and / or ML-guided parametric optimization.

[0078] The algorithm can comprise physical correlations for example material shrinkage behavior, heat transfer between contacts.

[0079] In a preferred embodiment of the inventive Method for generating custom connectors using an Al generator, it includes the step of aligning the physical requirement by a look up table, wherein aligning includes mapping the set of specification towards physical requirements using the look-up table.

[0080] Aligning the set of specification towards physical requirements using the look-up table can involve the step of mapping the user-provided requirements or the set of specification that are already defined. The look-up table is essentially a data structure, often used in computing, that can store known values in an organized way so that a program can quickly and efficiently find specific data. The look-up table can work as following.

[0081] Firstly, the look-up table can be configured to capture an array of possible specifications for different components or attributes of the product in question. This table would be indexed in such a way that it can quickly retrieve information based on a given input.

[0082] When a user provides their requirements to the LLM and the LLM would parse the input text to extract keywords, phrases, and numerical values that are relevant to the specifications as described above.

[0083] The parsed keywords and values are then matched against the keys or indices or value rages or threshold values in the look-up table. The matching process can involve some natural language processing (NLP) capabilities to handle synonyms, acronyms, or jargon typically used in the industry.

[0084] After identifying the correct keys in the look-up table, the corresponding specifications respectively the matching physical components of the connector can be retrieved. These retrieved values are the standardized specifications or components that match the user’s requirements or descriptions.

[0085] The look-up table respectively the Al generator then compiles all the selected specifications and presents them to the user in a coherent format, such as a report or a structured digital document or data set which can be provided to the generative design algorithm.

[0086] A final step of the Aligning step with the look-up table or the ML-guided parametric optimization can comprise human expert knowledge reviewing the output to ensure that all the requirements are met and the specifications are aligned with the user's needs and industry standards.

[0087] An example for the function of the look-up table can be read as following:

[0088] Human Input: "I need a connector that can handle at least 500 volts, is resistant to saltwater corrosion, and is suitable for outdoor marine environments."

[0089] Parsing with the LLM: The system identifies "500 volts," "resistant to saltwater corrosion," and "suitable for outdoor marine environments" as key requirements.

[0090] Look-Up Table:

[0091] "500 volts" -> Connector category: High Voltage, Spec: Rated Voltage > 500V

[0092] "resistant to saltwater corrosion" -> Material: Corrosion-resistant, Spec: Marine Grade Stainless Steel or Nickel-plated

[0093] "outdoor marine environments" -> Weatherproof Rating, Spec: IP67 or higher In a preferred embodiment of the inventive Method for generating custom connectors using an Al generator, it includes the step of aligning the physical requirement by an ML-guided parametric optimization, wherein the ML-guided parametric optimization makes use of a training dataset that includes various design parameters and their corresponding performance metrics so that the trained model can predict the optimal set of design parameters.

[0094] Machine Learning (ML)-guided parametric optimization uses advanced techniques such as regression, classification, and even deep learning models to predict and optimize parameters for a given set of requirements. This approach can be used as an alternative to a look-up table when designing custom components or products like electrical connectors. The ML process typically involves the following steps.

[0095] First, a comprehensive dataset that includes various design parameters and their corresponding performance metrics is assembled. This dataset might contain historical design data from previous connectors, simulation results, experimental data, or any combination thereof.

[0096] Then a machine learning model can be selected for the parametric optimization. The selection of an appropriate machine learning model for connector design optimization involves a systematic evaluation process that considers multiple factors including the nature of the design parameters, available training data, computational resources, and specific optimization objectives.

[0097] One efficient approach to model selection is utilizing AutoML (Automated Machine Learning) platforms available on the Internet, which can systematically evaluate multiple model architectures and hyperparameter configurations. They provide automated workflows that test various models against a test connector design dataset, including requirements as input and connector design as output. These platforms typically implement a range of algorithms including gradient boosting machines, neural networks, support vector machines, and ensemble methods, evaluating each based on performance metrics relevant to connector design optimization.

[0098] When selecting a model through AutoML or manual evaluation, several key criteria should be assessed. Prediction accuracy is paramount, particularly for critical parameters such as electrical performance characteristics and mechanical integrity. The model's ability to generalize across different connector types and use cases should be evaluated through cross-validation techniques. Computational efficiency becomes especially important when the model needs to be integrated into an interactive design workflow where real-time feedback is valuable. Interpretability of the model may be necessary when design engineers need to understand the reasoning behind specific parameter recommendations. Additionally, the model's ability to handle multi-objective optimization scenarios is crucial, as connector design often involves balancing competing requirements such as size constraints versus thermal performance.

[0099] For testing model candidates, a comprehensive evaluation protocol should be implemented. This includes splitting available connector design data into appropriate training, validation, and test sets, with consideration for stratification to ensure representative sampling across different connector types and applications. Performance metrics should be tailored to connector design requirements— for example, using mean absolute percentage error for dimensional parameters, classification metrics for categorical decisions like material selection, and custom metrics that reflect domain-specific constraints such as manufacturing feasibility or compliance with electrical standards.

[0100] Several model architectures have demonstrated particular promise for connector design optimization. For more complex optimization scenarios, deep learning approaches may be appropriate. Specifically, neural network architectures with specialized layers for handling both continuous parameters (dimensions, electrical ratings) and categorical features (material types, connector styles) can be effective.

[0101] Ensemble methods that combine multiple model types often yield superior results by leveraging the strengths of different approaches. For instance, a stacked ensemble might use random forests for initial parameter range estimation, followed by a more specialized neural network for fine-tuning specific dimensions. The model selection process should also consider the integration capabilities with existing engineering software and simulation tools. Models that can export their predictions in formats compatible with CAD systems, finite element analysis software, or electromagnetic simulation tools will streamline the overall design workflow. Additionally, models that support active learning— where they can suggest which design variants should be physically prototyped or simulated next to maximize information gain— can significantly accelerate the optimization process.

[0102] Ultimately, the selected model should be validated not only on historical data but also through limited physical prototyping or high-fidelity simulation of its recommendations. This final validation step ensures that the model's optimization suggestions translate effectively to real-world connector performance before being fully integrated into the production design workflow.

[0103] Once the model type, architecture and hyper parameters are selected, the ML model can be trained on the prepared dataset. The training process can involve using optimization algorithms to adjust the weights and biases within the model to minimize the difference between the model's predictions and the actual outcomes. Different ML models can be used for this purpose, such as decision trees for classification tasks, support vector machines for regression, or neural networks for complex relationships within the data.

[0104] For the training a certain number of previous connector designs can be used. Once the model has been trained, it needs to be validated and tested against a separate dataset not used during training to ensure that the model generalizes well to new, unseen data. An another number of previous connector designs can be used to check the raining Result of the ML guided parametric optimization.

[0105] The model can be refined by tuning its hyperparameters— settings that govern the model's architecture and learning process. Hyperparameter tuning is essential to find the ideal configuration that provides the best predictive accuracy.

[0106] Afterwards an identification of the most relevant features (design parameters) that significantly impact the performance can occur for example by manual input by comparing different designs and provide a rating to the different designs. Proper feature selection can improve the model's performance by excluding irrelevant or redundant data. The can also be done at the generative design algorithm.

[0107] For new requirements or specifications, the trained model can predict the optimal set of design parameters. This might involve running the model iteratively while adjusting the input requirements to optimize the desired performance metrics (such as minimizing weight while maximizing strength or electrical conductivity).

[0108] In another embodiment Al generator is configured to provide the 2d and / or 3d model set of data optimized set of information to a manufacturing station. The manufacturing station can then be configured to produce the respective custom connector. The manufacturing station can comprise at least turning machine and / or milling machine and / or injection molding machine and / or painting unit.

[0109] The output data of the optimized parameters for the connector may include a comprehensive set of specifications that define the connector's physical and electrical characteristics. This set may encompass dimensional information such as overall length, diameter, pin spacing, and contact depth; material selections for the conductor, insulator, and housing components; electrical specifications including voltage rating, current capacity, and impedance; environmental ratings like IP protection class; and mechanical properties such as mating force and durability cycle count. In some cases, the output may specify standard plug and socket configurations if applicable, or provide custom geometries for proprietary designs. The data set may also include tolerances for critical dimensions, surface finish requirements, and any special features like keying or locking mechanisms. Depending on the application, additional parameters such as operating temperature range, chemical resistance, or shielding effectiveness may be included. The specific dimensions and values for these parameters would vary based on the particular connector design and intended use, ranging from submillimeter precision for miniature connectors to centimeters for high-power industrial connectors.

[0110] Physics-based simulations may play a crucial role in validating that the optimized features of custom connectors fulfill user requirements. These simulations can provide detailed insights into the connector's performance under various conditions, allowing for comprehensive analysis even before physical prototyping. For electrical connectors, finite element analysis (FEA) may be employed to simulate electrical, thermal, and mechanical properties. In the realm of electrical performance, FEA can be used to calculate voltage distributions, current densities, and electric field strengths within the connector. This analysis may help in verifying that clearance and creepage distances meet safety standards, particularly important for high- voltage applications. The simulation can model the connector's geometry, material properties, and boundary conditions to predict potential areas of electrical stress concentration or breakdown. Thermal simulations using FEA techniques may be applied to assess heat dissipation characteristics. These simulations can model the heat generated by current flow and predict temperature distributions across the connector under various operating conditions. This analysis may be particularly valuable for high-current applications where thermal management is critical. Mechanical simulations using FEA can evaluate the connector's structural integrity under various loads, including insertion and extraction forces, vibration, and shock. These simulations may help in predicting stress concentrations, deformation, and potential failure points, ensuring that the connector meets durability and reliability requirements. Additionally, multiphysics simulations that couple electrical, thermal, and mechanical analyses may provide a more comprehensive understanding of the connector's behavior, accounting for phenomena such as thermal expansion and its effects on electrical contact resistance.

[0111] The results from these physics-based simulations can be used to iteratively improve the optimized features of the connector design. This iterative process may involve a feedback loop where simulation outcomes inform design modifications, which are then re-simulated to assess their impact. For instance, if thermal simulations indicate localized hot spots, the design may be adjusted to enhance heat dissipation in those areas, perhaps by modifying the connector's geometry or material composition. The modified design can then be resimulated to verify improvement. Similarly, if electrical simulations reveal areas of high field strength, the design may be refined to redistribute the electric field more evenly. This iterative approach may also involve sensitivity analysis, where key design parameters are systematically varied to understand their impact on performance metrics. Machine learning techniques may be integrated into this process to guide the exploration of the design space more efficiently. For example, surrogate models or metamodels trained on simulation data can rapidly predict performance for new design variations, allowing for faster iteration cycles. As the design evolves, increasingly detailed and computationally intensive simulations may be employed to fine-tune specific aspects of the connector. This iterative simulation-driven optimization process may continue until the design meets or exceeds all specified requirements, balancing various performance criteria and manufacturing constraints. The final optimized design may then be validated through a comprehensive set of simulations before proceeding to physical prototyping and testing.

[0112] Providing the optimized parameters in a readable and understandable format for engineers or stakeholders or the generative design algorithm.

[0113] An example scenario in which the ML-guided parametric optimization can be used instead of the look-up table might be when designing a custom connector where the ideal specifications (such as dimensions, material, number of pins, electrical ratings, etc.) need to be balanced against a set of constraints (like cost, weight, environmental factors, etc.). The ML model would process the requirements and, by recognizing patterns in the data, propose the optimal design that meets the desired criteria.

[0114] This approach can be more flexible and scalable than the look-up table, particularly for complex optimization tasks where there are many interacting factors. ML models can handle nonlinear relationships between parameters and outcomes, and they can adapt to new data as designs and technologies evolve, whereas the look-up table is static and may not capture the complexity or the interdependencies between design variables. However, the effectiveness of ML-guided optimization largely depends on the quality and quantity of the training data and the appropriateness of the selected ML approach for the specific engineering problem at hand.

[0115] Generative design with the generative design algorithm is a method used in computer-aided design (CAD) that can leverage algorithmic and computational processes to generate a wide array of design alternatives based on user-defined goals and constraints. Optimization algorithms can be used, often including machine learning techniques to explore the possible design space and create optimized 2D or 3D models respectively a 2d and / or 3d set of data tailored to the set of specifications and the physical requirements.

[0116] The generative design algorithm can be or comprise an iterative process and / or step. After providing a first 2d and / or 3d set of data. The 2d and / or 3d set of data can be rated by an evaluating step. The optimization generative design algorithm can also be configured to internally iteratively improve the output of the 2d and / or 3d set of data.

[0117] The Iterative step can comprise to evaluate certain thresholds and / or specifications and returning information to the generative design algorithm to further refine the first 2d and / or 3d set of data or before outputting the first 2d and / or 3d set of data.

[0118] The generation process starts with the specifications and the physical requirements that describe the desired characteristics of the final design. These might include constraints such as material properties, strength, weight, cost, manufacturing methods, and environmental conditions, as well as any specific design variables to be considered.

[0119] The generative design algorithm can use the input data to explore the possible design space. It typically can generate many iterations of potential designs by varying the defined design variables within the given constraints. This iterative process is done rapidly and can explore more options than a human designer typically would.

[0120] Each design can be evaluated using simulations to determine how well it meets the specified goals. The algorithm which can be used is a finite element analysis (FEA) to assess strength, durability and / or electrical properties, computational fluid dynamics (CFD) to evaluate airflow or fluid movement, thermal analysis to check for heat resistance, and other parameters.

[0121] The algorithm can apply optimization techniques to refine the designs. It can learns from each simulation, identifying which parameters lead to better outcomes and adjusting accordingly. Effective generative design may also employ evolutionary algorithms or other machine learning methods, which can mimic natural selection by retaining 'successful' designs and iterating upon them. A suite of the best-performing designs can be presented to the user, often showing a variety of solutions that balance the trade-offs between different criteria of the set of specifications and the physical requirements in different ways.

[0122] A user can review the generated designs and can either select a final design, adjust the input parameters and constraints to refine the results, or even combine elements from different designs to create a hybrid solution.

[0123] Once the user is satisfied with the outcome, the chosen design can be further detailed and prepared for prototyping or production, including generating the necessary 2d and / or 3d model set of data.

[0124] In an embodiment the physical requirements comprise at least one of the following parameters:

[0125] • Type of Connector comprising the standard or specific design of the connector, such as

[0126] BNC, SMA, N-Type for RF connections, or USB, HDMI, RJ45 for data and audio / video applications;

[0127] • Gender, wherein connectors can be typically described as male (pins) or female (sockets), indicating how they mate with corresponding components.

[0128] • Contact material, comprising the material which can be used for the conducting part of the connector, which might be gold, silver, or nickel-plated to ensure good conductivity and corrosion resistance.

[0129] • Insulator Material, which can be the non-conducting part that provides the necessary electrical isolation between pins or between the connector and the housing. Materials typically include plastics like PTFE (polytetrafluoroethylene) or durable polymers.

[0130] • Size and Dimensions can comprise the physical dimensions including diameter, length, and profile, which are crucial for ensuring the connector fits into the corresponding device or port; for RF connectors, this also can include the interface dimensions defined by standards.

[0131] • Contact Number can comprise the number of individual conducting elements within the connector which can be configured to carry a signal or power.

[0132] • Rated Voltage and Current can comprise the maximum voltage and current that the connector can safely handle. • Impedance (for RF connectors) can comprise the characteristic impedance of RF connectors, which can be 50 or 75 Ohm for most applications, is critical for maintaining signal integrity and minimizing reflections.

[0133] • Frequency Range (for RF connectors) can comprise the range of frequencies over which the connector can perform well, causing minimal loss and distortion.

[0134] • Durability can comprise This refers to the number of mating cycles a connector can withstand while still maintaining its mechanical and electrical integrity.

[0135] • Environmental Resistance can comprise the ability of the connector to withstand environmental conditions such as temperature extremes, moisture, UV exposure, and chemical exposure.

[0136] • Cable Attachment Mechanism and Strain Relief can comprise how the connector attaches to and provides support for the cable to prevent damage due to bending or pulling.

[0137] • Interface Standards can comprise that multimedia connectors might adhere to standards like

[0138] HDMI, DisplayPort, or USB-C, which can carry audio, video, and data simultaneously.

[0139] • Bandwidth can comprise for connectors carrying high-definition video or high-speed data, the bandwidth indicates the maximum rate at which data can be transmitted.

[0140] • Audio / Video Channel Support can comprise the number and type of audio and video channels supported (e.g., HDMI connectors support multiple audio channels and 4K video).

[0141] • Data Transfer Rate can comprise the maximum speed at which data can be transferred through the connector, relevant for USB or Thunderbolt connectors.

[0142] • Support for Power delivery can comprise that some connectors, like USB-C, can also deliver power, so the rated voltage and current for power delivery are important aspects.

[0143] • Digital Rights Management (DRM) can comprise that some connectors transmitting protected content, support for DRM may be necessary (e.g., HDMI connectors support HDCP for protected content transmission).

[0144] • Signal Conversion or Adaptation can comprise that some multimedia connectors incorporate active or passive components to convert from one signal type to another or to adapt impedance.

[0145] • Physical Configuration can comprise whether the connector is straight, angled, or rotatable, which can be important for cable management and equipment placement.

[0146] • Locking Mechanism can be integral to multimedia connectors, ensuring a secure connection that will not be accidentally disconnected.

[0147] Color-coding can comprise that Multimedia connectors might use color-coding to aid in identifying the correct port for connection, especially in complex setups. • EMI / RFI Shielding can comprise shielding which is essential to prevent electromagnetic interference (EMI) and radio frequency interference (RFI) that can degrade signal quality.

[0148] In an embodiment the generative design algorithm uses a derivation function for enhancing its ability to quickly converge to optimal or near-optimal solutions from the space of all possible designs.

[0149] Accelerating a generative design algorithm can involve enhancing its ability to quickly converge to optimal or near-optimal solutions from the space of all possible designs. One way to achieve this can be by using a derivation function, which in the context of optimization algorithms, can refer to the mathematical process of finding the gradient or derivative of a function. This approach is largely related to gradient-based optimization methods.

[0150] The derivation function also called gradient-based approach computes the gradients (partial derivatives) of the objective function with respect to the design variables. The objective function specifies what you're trying to optimize (e.g., minimize weight, maximize strength).

[0151] The gradient can represent the direction of the steepest increase of the objective function. For optimization, you're interested in the opposite direction (the steepest decrease) if you're minimizing, or the direction as is if you're maximizing.

[0152] The algorithm then can use this gradient information to update the design variables, pushing the design towards an improved solution. This update might involve a simple step in the opposite direction of the gradient (in the case of minimization) or require more sophisticated techniques like line search or trust region methods to choose an appropriate step size.

[0153] By leveraging the gradient information, generative design algorithms can make informed, efficient steps towards the optima rather than taking random or exhaustive search steps. However, the direct use of derivation functions assumes that the objective function is differentiable with respect to the design variables. This makes the method using a derivation function more effective.

[0154] If a derivation function is applied effectively, it can significantly accelerate the convergence of the algorithm by guiding the search process through the design space more intelligently. Sometimes derivative-free methods such as evolutionary algorithms can be combined with gradient-based approaches, taking advantage of both robustness to landscape features and efficient gradient-guided refinement.

[0155] The invention also provides a product data file generated using the Al generator according to at least one of the previous embodiments.

[0156] The invention also provides a computer program product that can be loaded into a computer memory of a device according to the Al generator according to at least one of the previous embodiments.

[0157] The invention also provides a computer system configured to execute the Al generator according to at least one of the previous embodiments.

[0158] The invention also provides a method for producing, comprising the steps of:

[0159] Executing the method according to one of the previous embodiments;

[0160] Producing a connector according to a connector model generated in step 1 .

[0161] In another embodiment a finite element model validity check is applied to the 2d and / or 3d custom connector model for plausibility check.

[0162] In an embodiment the finite element model validity check outputs a check information if the 2d and / or 3d custom connector model complies with a set of mechanical specifications.

[0163] The invention is not limited to the described embodiments. Within the scope of the invention, all described and / or drawn features can be combined with each other as desired, unless otherwise indicated.

[0164] The invention is described below by way of example only, with reference to the drawings. It shows: Figure 1 a method for using an Al generator for generating custom connector.

[0165] Figure 1 discloses a method for using an Al generator for generating a custom connector. The Method comprising the following steps. Providing some input information to a LLM (S1) wherein, the input information comprises human natural language information. Processing the input information with the LLM into a set of specification that is interpretable by a computer (S2). Aligning the set of specification towards physical requirements (S3).

[0166] Using a generative design algorithm for generating an optimized 2d and / or 3d model set of data incorporating the set of specifications and the physical requirements (S4). Outputting a 2d and / or 3d set of data provided by the custom connector model (S5).

[0167] Reference signs

[0168] Steps S1 - S5

Claims

Claims:1 . Method for generating custom connectors using an Al generator, comprising the steps of:- Providing some input information to a LLM wherein, the input information comprises human natural language information (S1);- Processing the input information with the LLM into a set of specification that is interpretable by a computer (S2);- Aligning the set of specification towards physical requirements (S3);- Using a generative design algorithm for generating an optimized 2d and / or 3d model set of data incorporating the set of specifications and the physical requirements (S4);- Outputting a 2d and / or 3d set of data of the custom connector model (S5).

2. Method for generating custom connectors using an Al generator according to claim 1 , wherein the LLM is configured to validate the input information according to at least one of the following parameters:• value range;• threshold value;• sign;• contradictions.

3. Method for generating custom connectors using an Al generator according to at least one of the previous claims, wherein the Al generator is configured to be executed on premise or to outsource tasks to a cloud if the capacity utilization exceeds a certain threshold, wherein the threshold takes into account at least one of the following metrics: percentage oftotal available CPU utilization, GPU memory consumption, RAM usage, number of floating-point operations per second, total duration of running a task.

4. Method for generating custom connectors using an Al generator according to at least one of the previous claims, comprising the step of- outputting the processed information from the human natural language information comprising the set of specification to a user from the LLM.

5. Method for generating custom connectors using an Al generator according to at least one of the previous claims, comprising the step of- aligning the physical requirement by a look up table, wherein aligning includes mapping the set of specification towards physical requirements using the look-up table.

6. Method for generating custom connectors using an Al generator according to at least one of the previous claims, comprising the step of aligning the physical requirement by an ML-guided parametric optimization, wherein the ML-guided parametric optimization makes use of a training dataset that includes various design parameters and their corresponding performance metrics so that the trained model can predict the optimal set of design parameters.

7. Method for generating custom connectors using an Al generator according to at least one of the previous claims, comprising the step of- providing the optimized 2d and / or 3d model set of data to a manufacturing station.

8. Method for generating custom connectors using an Al generator according to at least one of the previous claims, wherein the physical requirements comprising at least one of the following parameter:• Type of Connector, Gender, Contact Material, Insulator Material, Size and Dimensions, Contact Number, Rated Voltage and Current, Impedance, Frequency Range, Durability, Environmental Resistance, Cable Attachment Mechanism and Strain Relief, Interface Standards, Bandwidth, Audio / VideoChannel Support, Data Transfer Rate, Support for Power delivery, Digital Rights Management (DRM), Signal Conversion or Adaptation, Physical Configuration, Locking Mechanism, Color-coding, EMI / RFI Shielding.

9. Method for generating custom connectors using an Al generator according to at least one of the previous claims, wherein the generative design algorithm uses a derivation function for enhancing its ability to quickly converge to optimal or near-optimal solutions from the space of all possible designs.

10. Product data file generated using the Al generator according to at least one claims 1 to 9.11 . Computer program product that can be loaded into a computer memory of a device according to the Al generator according to at least one claims 1 to 9.

12. Computer system configured to execute the Al generator according to any one of the claims 1 to 9.

13. Method for producing a custom connector, comprising the steps of:Executing the method according to at least one of the claims 1 to 9;Producing a connector according to a connector model generated in step 1 .

14. Method for producing a custom connector, according to claim 13, wherein a finite element model validity check is applied to the 2d and / or 3d custom connector model for plausibility check.

15. Method for producing a custom connector, according to claim 14, whereinthe finite element model validity check outputs a check information if the 2d and / or 3d custom connector model complies with a set of mechanical specifications.

Images