Hackathons to Accelerate the Development of Low-Cost Digital Solutions

ABSTRACT Digital developments undertaken as part of the academic research projects can be slow to be trialed and adopted industrially. These delays may be the result of lack of suitable early prototypes or insufficient industrial workforce engagement among other challenges. Here, these challenges are considered in the context of a research project that develops procedures for the deployment of simple, low-cost digital solutions by small manufacturers. This paper proposes a methodology that comprises the use of hackathons, laboratory demonstrators, and industrial pilot activities as a means of accelerating the use of solutions developed. In particular, it introduces hackathons as a means to quickly create cost-efficient digital solutions for manufacturing and argues how, in addition to speeding adoption, these could be seen as additional option to help prepare the future workforce.


Introduction
Academic research projects often face delays when trying to progress innovative digital developments. For instance, developments could be slow to be trialed and adopted industrially. These delays may be the result of lack of suitable early functional prototypes key for demonstrating proof-of-concept ideas, illustrating progress, and trialing support pilots in small and mediumsized enterprises (SMEs), or insufficient industrial workforce engagement among other challenges. Employing open and social innovation activities could be an option to accelerate the production of deliverables. Among these, hackathons, which are sprint-like design and development events, have been considered as a mechanism for rapid innovation in a variety of domains. 1 Seen as controlled ecosystems where groups of individuals cooperate in teams to quickly achieve something beneficial, one of the reasons hackathons are successful is because participants come from different backgrounds, hence contributing and sharing diverse expertise and skills while identifying themselves with solution end-users. 2 However, since there are different types of hackathons the interest is put on addressing the following question: is it possible to define a hackathon where participants are given requirements observed in manufacturing SMEs and produce functional prototypes capable to demonstrate the benefits of low-cost digital solutions?
Thus, the first aim of this work is to introduce a new type of technical hackathon, called "Shoestring hackathon," focused on creating cost-efficient prototype solutions that solve digital needs observed in manufacturing SMEs. In particular, these solutions follow a modular design and employ low-cost, off-the-shelf, open source technologies from both industrial and non-industrial domains. Thereby, having the ability to build proof-ofconcept digital solutions in a very short space of time is helpful for demonstrating potential. However, in order to encourage manufacturing SMEs adopt digitalization, such prototypes might need specific enhancements, technical tuning and tailoring to operate effectively within industrial environments. Hence, a pathway where developments go from requirement to industrial pilot is required. The research is then centered on the following question: would it be possible to define a methodology that helps accelerate the design of lowcost digital solutions, the creation of fully functional prototypes, and the realization of pilots for industrial digitalization?
The second aim of this paper is then to propose a methodology for the rapid development of low-cost digital solutions for manufacturing SMEs. This process comprises elements of a structured digitalization approach, such as Digital Manufacturing on a -Shoestring, 3 Shoestring hackathon solutions, demonstrators development (i.e. lab environment improved prototypes), and pilot activities (i.e. tailored solutions developed and evaluated in industrial environments). As a result of adopting digital technologies, changes occur not only in manufacturing production environments but also in the way people work. Therefore, novel strategies are needed to address the mismatch between the digital skills required by industry and the workforce capabilities. Thus, there is great potential for manufacturing and educational organizations to shape and develop the future workforce with digital knowledge and skills. The third aim of this paper is to present a brief discussion based on the following question: could Shoestring hackathons be additional options to help prepare the future workforce, e.g. in further education colleges or apprenticeship organizations, and help upskill the current workforce, i.e. across shopfloor operatives?
First, this paper sets the context for low-cost digital solutions and discusses the needs for accelerating innovation. Then, it proposes an approach for rapid development of low-cost digital solutions, illustrates this by presenting two representative industrial cases, discusses hackathons as a mechanism for developing digital skills and, finally, it summarizes the achieved results.

Background
In this section, we review the relevant literature. First, we highlight the importance of developing low-cost digital solutions for manufacturing SMEs. Second, we look strategies and methodologies used to accelerate the delivery of digital solutions. Finally, we revisit some examples of rapid innovation via hackathons and the types of hackathons.

Developing low-cost digital solutions for small manufacturers
The low adoption level of digital technologies such as artificial intelligence, internet of things or data analytics is one of the key barriers to industrial digitalization among UK manufacturing SMEs. In particular, it has been reported that a relatively small number of these SMEs explore the adoption of advanced digital technologies in their manufacturing process 4 and that digitalization priorities are common among many SMEs. 5 This sheds light on a real need for making digital solutions more accessible to manufacturing SMEs who are often less able to adopt digital solutions due to perceived high investment costs, lack of digital skills in the workforce, the need to integrate with legacy systems, and the need for solutions that are simple to deploy.
In an effort to increase accessibility, the first steps toward developing a more elaborate process that helps manufacturing SMEs decide on cost-efficient digital implementations is proposed by the Digital Manufacturing on a Shoestring (DMS) approach. 6 This proposes a problem-centric view on digitalization through the use of modular elements that offer an incremental solution design procedure and the adoption of low-cost, off-the-shelf, open source industrial and nonindustrial digital technologies. The resulting digital solutions from this approach are typically add-on, peripheral systems capable of working independently from existing core manufacturing processes. Contrary to most commercial solutions which are usually vendor locked, overfitting and expensive developments, these cost-efficient solutions are not meant to be permanent deployments, safe critical, fully automated, work unsupervised, control critical operations or require high accuracy.
To this end, the DMS approach involves five main steps: (1) investigating and prioritizing the digital solutions that manufacturing SMEs need most, (2) developing lab demonstrators and industrial pilots for those solutions, (3) reviewing the viability of developed solutions with input from industry and academia, (4) training end-users and (5) offering general operational and maintenance procedures where possible. It is the second step where low-cost is ensured by using off-the-shelf, non-industrial hardware and software to address a company's (digital) solution needs. Developed demonstrators and pilots seek to exploit inexpensive commercially available technologies such as mobile computing, sensing, micro processing, visualization, and analytics while addressing the challenges associated with integrating these safely and securely into manufacturing environments. Developing digital solution demonstrators is important as they can be used to inspire and engage industry or further expand these through integration into real manufacturing production environments which contributes to the early industrial assessment proposed in the third step.

Options for accelerating digital solution development
There are many known strategies that have been used to accelerate the delivery of project outcomes. Crowdsourcing, for instance, can reduce both development costs and time to market while improving product quality, flexibility, scalability, and diversity. 7 However, the outcomes of crowdsourcing activities could often be influenced by both monetary motivations of the leader of the initiative (whether a person or a company) and the pool of participating end users (the crowd). These parties can bias outcomes to suit their individual benefits and interests which increases the delivery risks.
Moreover, the developed outcome could end up with a highly specific function but lack broad appeal if the initiative leader has a strong vision, or if the project is supported by a small crowd of well aligned enthusiasts. Alternatively, it could struggle to accommodate the varying (and potentially conflicting) requirements from stakeholders and thereby take longer to deliver. 8 Furthermore, crowdsourced projects face organization and operational barriers that arise from misalignment between budgeting and project timelines, unclear responsibility for managing and validating outcomes, and a lack of predefined process structures. 9 Brainstorming is a time-honored collaborative option widely used for over 70 years across organizations from different domains and sectors. However, this popular method has proven to have several limitations such as freeriding, regression to the mean, evaluation apprehension, and creativity blockage. 10 Thus, reducing the chance to offer a fertile ground for innovation, creativity, and co-creation to flourish. For these reasons, brainstorming is an effective technique when used for generating ideas, i.e. right before embarking into any design, prototyping, or testing stage, however in other scenarios its usefulness can be limited. 11,12 Other more tailored methodologies, such as the theoretical framework for digital innovation, have been proposed in recent years. 13 The framework considers key phases of the innovation process such as initiation, development, implementation, and exploitation as well as other key factors such as the internal organizational environment, the external competitive environment, and the type of outcome (whether it is a product, process, or service). Although this proposes a structured approach, its focus on offering a digital innovation process to follow is limited by the lack of a case study or actual industrial application to consider. It is therefore unclear if it addresses challenges arising from reduced workforce or its potential impact for accelerating development.

Achieving rapid innovation via hackathon events
The realization of proof-of-concept prototypes can be hindered by internal and external barriers within an organization. Common internal barriers are: work forces with restrictive mind-sets, poor knowledge of employee competences and skills, having an organizational culture that is resistant to innovation and continuous improvement, and lack of access to information on relevant technologies. 14 To overcome these, incremental development approaches, such as agile or lean methods, have been developed. Unfortunately, these methods were found to focus on efficiency and meeting requirements, and tend to overlook emergent creativity and thinking 15 as well as other temporal structures useful for accelerating innovation. As an alternative, organizations have been moving from inward focused innovation toward open innovation with the objective of disrupting internal processes and encouraging the use and exchange of external ideas, technologies, knowledge, talent, and resources. 16 Thus, one of the many ways to achieve this is through co-creation events like hackathons. These can be defined as flexible invention development methods where attendees face specific challenges to solve within a limited amount of time. 10 The use of hackathons for rapid innovation has been adopted in a variety of domains. For instance, in a 36 hours hackathon organized to address six IoTbased challenges, every team was given an educational IoT development kit, valued less than £100, together with access to cloud computing services and a bespoke IoT template infrastructure designed to rapidly instantiate a specific architecture for digital solutions. 17 As a result of the event, one of the sponsoring companies considered using a digital solution to upgrade their legacy refrigerators. This was a cost-effective, innovative, add-on system to monitor fridge operating conditions, record amount and shelf life of stored products, and automatically generate online orders on-demand. The system is based on weight, temperature, and humidity sensor readings and built using Arduino microcontrollers, Raspberry Pi microcomputers and the IBM Bluemix platform.
Another example of rapid innovation from hackathons is observed in the glass tempering industry. 18 In particular, the challenges presented in this competition were derived from glass processing problems scooped by a glass manufacturing company called Glaston. In response to this call, a total of 72 participants including startups, engineering students and researchers teamed up, developed, and then pitched solution ideas. The winning team proposed a consumer-centric, big-data analytics solution in the form of a mobile application. Glaston sponsored further academic research followed by a practical implementation and, approximately one year later, a mature digital solution was released combining both artificial intelligence and augmented reality to analyze glass fragmentation patterns from images. This was made public as the Glaston Siru app which supports both European and American standards.
In an effort to understand how accelerated innovation emerges in hackathons, the conditions and outcomes from this type of events were systematically studied across 13 challenges. 19 Out of all the proposed solutions, it was observed that 7 offered no functionality, 3 offered a basic level of functionality and 3 were fully functional. Despite similar conditions, it emerged that there was a variation in how participants dealt with temporal ambiguity, i.e. the time frames, sequence, progression, and length of activities needed to produce a solution. For instance, all the teams that delivered nonfunctional solutions imported organizational structures from past experience (e.g. agile development processes) in an attempt to fully coordinate team work. In contrast, the other six teams started with the assumption that accelerated innovation was different and let new organizational structures emerge to form a minimal basis for coordination while working in an adaptive manner.
There are a vast number of examples from different application domains using hackathon events as pathways for rapid innovation. Across them, and as noted in the three cases mentioned above, hackathons have been characterized as "unique" in the level to which they provide freedom to improvise and self-organize while enhancing creativity and requiring both convergent and divergent thinking. 10 From these features emerges a foundation of flexibility needed to deal with ad-hoc time frames without underlying temporal organizational structures while offering participants the chance to exploit nonhierarchical and open ways of organization. 19

Hackathons for accelerated digital solutions
Some of the many ways to classify hackathon events are according to their type, approach, and competitive manner. The first characteristic classifies a hackathon according to the domain and background of the participants, the second considers how the event is conducted and what participants are expected to deliver while the third considers the level of competitiveness. The hackathon type can be divided into: the classic type, where participants have a technical background and focus on achieving a technical solution for a challenge; the free type, where participants contribute with different skill sets; or the mixed type, where teams are a blend of the first and second types. The hackathon approach can be classified as: an open approach, where participants are free to address any challenge they want around a given topic and expected to provide as many candidate solutions as they want; a closed approach, where the challenges are predefined and participants are expected to provide a fixed number of solutions; or a dedicated approach where there is a single challenge for which participants are expected to deliver candidate solutions. While hackathons can be classed according to many other aspects, e.g. their competitive manner, there are systematic studies with in-depth characterizations to consider when choosing the type of hackathon to organize. 20,21 Regardless of the chosen type of hackathon, it is yet recognized that these events encourage the production of demonstrator systems, while acknowledging that subsequent technical work is always required to realize a complete and robust product. 22 In summary, there is a need to address the barriers to industrial digitalization among UK manufacturing SMEs by making digital solutions cheaper and simple to deploy. Compared to other strategies, it has been shown that hackathons are successful to accelerate industrial digital innovation. However, since the outcomes of hackathon events are not complete products, organizations still need to consider pathways and outreach activities to transform the resulting prototypes into ready-to-use products. 20 Therefore, these represent an opportunity to characterize and establish a practical approach that, within digital research context, goes from problem specification to solution development with industrial adoption. The rest of this work presents a hackathon-driven solution development pathway to ensure digital research developments are moved quickly into industrial use.

A hackathon driven digital solution development approach
The following presents an approach for accelerating the development of low-cost digital ("Shoestring") solutions for manufacturing SMEs by using hackathons to streamline the design and development of minimal, costefficient, digital prototypes. From this point on, we will refer to hackathons developed for this purpose as "Shoestring hackathons."

The shoestring hackathon
A Shoestring hackathon is defined here as a "technical development hackathon characterized by a set of industrial digital challenges and the features of the resulting digital solution prototypes." In particular, the challenges set to participants originate from common digitalization priorities observed across UK manufacturing SMEs. 5 The prototypes resulting from a Shoestring hackathon are designed with the modular elements offered by the DMS approach 6 and realized using low-cost, off-theshelf, open source software and hardware digital technologies from both industrial and non-industrial domains. [23][24][25] As a secondary objective, such hackathons are meant to highlight the existence of industrial challenges to open source or "hacker" communities who may be interested in contributing to the production of minimal solution prototypes, identifying and collecting low-cost off-the-shelf (sometimes newly emergent) technologies, fostering their own hands-on digital knowledge and abilities in preparation for working in digitized industrial environments as well as assessing the methodology offered by the DMS approach.

The hackathon driven digital solution development process
The methodology proposed here aims at defining a pathway to help accelerate digital research into pilots for industrial adoption. This is a qualitative approach that can be broadly divided into three steps (see Figure 1). The principle behind these enable: rapid development of digital solution alternatives to a given need, raise awareness that cost-efficient industrial digitalization is possible and help manufacturing organizations seamlessly adopt digital solutions.
The first step takes place within a Shoestring Hackathon Environment. This is where a Shoestring hackathon event is organized with challenges scoped from a prioritized list of digital solution areas. During such technical competition, teams of digital practitioners design and implement digital solution prototypes that solve the proposed challenges. In particular, the design follows the incremental solution design procedure from the DMS approach and the implementation employs low-cost technology kits such as sensors, micro-computers, micro-controllers, breadboards, etc. provided by the event organizers.
The second step takes place within a Lab Environment. This is usually a laboratory or workshop where the hackathon solution prototypes are technically inspected and evaluated, and the best of them are converted into demonstrators. In particular, the code is cleaned and optimized, boards and wiring are secured and housed, and some technologies get replaced to improve functionality. A demonstrator is not only a technically enhanced solution prototype, but also a reconfigurable and adaptable proof-of-concept system built to highlight specific digital capabilities and their potential to add value to certain manufacturing SME operations (examples at https://www.digitalshoestring. net/the-shoestring-process/shoestring-demonstrators/). The third step takes place within an Industrial Environment. This is usually a manufacturing SME site where demonstrators that the organization is interested to trial are converted into industrial pilots. In particular, the digital solutions are configured and tailored according to business needs and industrial conditions. For example, user interfaces are restructured to match the organization layouts while housings are ruggedized and secure to comply with shopfloor requirements. Pilots are peripheral to core manufacturing operations and deployed as independent digital solutions with the help of on-site staff who is also trained to operate and carry out their maintenance (examples at https://www.digital shoestring.net/the-shoestring-process/case-studies-anddemos/#PILOTS).

Stage 1: hackathon bared solution development
Prior to arranging a hackathon, the process starts with the identification of priority digital solutions needed by many manufacturing SMEs. These are scoped from a catalog of existing digital solution areas and specified as structured hackathon challenges within which a prototype design and development can be delivered in two days. These challenge specifications include required datasets, development kits, and input parameters. In this way, the outcomes of a Shoestring hackathon are quickly built prototypes that prove concepts and validate ideas. Technically speaking, the prototypes resulting from a Shoestring hackathon are typically functional solutions implemented with inexpensive digital technologies "stitched" together to work under very specific set of conditions established by the participants. These artifacts are not expected to be in an industrially usable state at this point and will likely have rushed and undocumented code, frail fabrications, inefficient algorithms, loose wires, or breadboard connected electronics.

Stage 2: laboratory proofing
The second stage of the process takes place in a lab environment. In this phase, the hackathon prototypes are collectively inspected, tuned, profiled, and tested before they are turned into demonstrators through further technical enhancement. These improved solutions are then showcased across a variety of engagement activities with the goal of raising awareness about the potential of low-cost digital solutions and getting input from observers about unconsidered applications or domain needs. Industrial engagement activities are opportunities to capture the interest of businesses and, as a result, identify which demonstrators could be carried forward as pilot deployments.

Stage 3: industrial piloting
In the third stage of the process, the identified demonstrators are further tailored and prepared for use in an industrial environment. In particular, the operational context of the solution within a manufacturing SME is identified and considered to determine where it would be used, by whom and whether there are any environmental considerations that need to be accounted for in the design, e.g. would the digital solution operate on the shopfloor or would it provide management support from an office area. Part of this process includes making any necessary adaptations, assessments, and evaluations to enable safe and secure integration into the production environment, and ensuring adherence to any relevant legislation, or technical standards employed by the organization. Thus, the deployment of a pilot solution results in a complete operative solution exemplar that is tested and affirmed. As a consequence, industrial case studies are conducted to record the benefits brought to the business, e.g. added values, contributions, benefits, and improvements.

Case studies
This section presents the application of the hackathon driven digital solution development process. In particular, it describes and evaluates two applications that have been specifically chosen because their technical characteristics are representative of two digital solution areas, i.e. process monitoring, and automated delivery of workpieces and tools to operators, 5 and because they illustrate how different low-cost and off-the-shelf technologies can be combined.

Student hackathons
Three hackathons addressing manufacturing SMEs digitalization problems as challenges were conducted in October 2019, October 2020 and February 2022. Out of the 34 solutions developed for the eight challenges proposed across the three events (see Table 1), the best were collectively inspected, tested, and further enhanced to turn them into lab demonstrators. This not only complied with the outputs  required by the underlying program, but also accelerated the progress needed to validate ideas. Although the following subsections present the application of the proposed approach, the links in the table below provide more information about the events and the challenges.

Case 1: legacy panel data extraction
The Legacy Control Panel Status challenge is a representative case of a digital solution that captures live operational data in real-time, triggers alerts, and stores past data for subsequent analysis and improvement of production planning. The solution developed for this requirement underwent the three stages of the hackathon driven digital solution development process (see Figure 2). In each of these, different levels of readymade, off-the-shelf, low-cost digital technologies were put together with the objective of demonstrating a proof-of-concept solution idea, showcasing potential benefits and capabilities for manufacturing environments, and enable industrial integration for real-world evaluation.

Stage 1: hackathon solution
The best solution resulting from the Legacy Control Panel Status challenge was a vision system consisting of a USB camera connected to a microcomputer equipped with an image processing module and a notification module implemented with open source Python libraries. All these at a total cost of just under £60 (see Figure 3). This prototype operates by facing the USB camera toward the front of a preprogrammed mock control panel (provided to the participants) which has panel elements representative of legacy industrial equipment (e.g. knobs, LCD displays, switches, and analogue dials). The camera continuously feeds a video stream into the image processing module which is configured to sample video frames at a specific rate. These frames are then processed in turn to identify and classify status change. For example, measuring the pixel intensity to decide whether a green light is switched on or off. Likewise, a template-based method is used to recognize the information of a seven segment LCD display while the angle of the needle is used to read the dial. Thus, when a change in the mock panel takes place, the image processing module automatically outputs the identification of the changing element and its newly associated status. This information is then automatically uploaded to a cloud-based spreadsheet and made available for visualization while the notification module delivers a panel status update via e-mail.

Stage 2: demonstrator
The development of a solution demonstrator focused on a series of activities defined to improve the accuracy of the hackathon solution, test the software, and evaluate different hardware alternatives. In order to improve the accuracy and efficiency, a machine learning (ML) module was trained to identify and classify status change of the seven segments LCD display. Thus, when a change in the panel takes place, both the image processing module and the ML module automatically output the identification of the changing element and its newly associated status. A number of functional procedures were designed with the objective of comparing between a deep learning model and a k-nearest neighbor (kNN) model, and also to make quantitative assessments of the results obtained by the vision system as a whole. For each functional procedure, the mock panel was programmed with an arbitrary set of instructions to automatically change the status of its elements. Once an instruction was executed, key performance indicators, i.e. the identification of the changing element and its newly associated status together with the overall processing time, were captured and compared against the actual updates within the mock panel. The functional procedures were executed in turn with the aim of improving overall identification accuracy (i.e. ensure panel elements and their new status are correctly identified), as well as to minimize response time to changes (i.e. guarantee that the image processing module and the ML module have an acceptable execution time).
Consequently, the kNN model was chosen against the deep learning one because it scored similar performance and required less data for training. In order to test the software, the solution was left to run continuously for 48 hours with the goal to validate, profile, and debug the implementations of the image processing module, the ML module and the notification module. As a result of this activity, more efficient, reliable, and faster software implementations were achieved. The evaluation of different hardware alternatives involved the use of different lighting environments, cameras, and software libraries. Thus, resulting in a solution which is simpler to adapt and configure to different industrial needs. Figure 4 depicts the demonstrator development sitting on a lab bench for accuracy and reliability improvement during a functional procedure.

Stage 3: pilot
The manufacturing SME interested in the solution demonstrator was a medium-sized company that uses equipment controlled by legacy panels which indicate to shopfloor staff the current operational status using lights, dials, and seven-segment numerical displays. In particular, the company wanted to enhance their existing manufacturing processes by automating the capture and collection of the current operational status of the machinery and provide real-time access to it. Hence, they needed to convert the panel's visual variables readout into digital data without modifying any processes or tampering equipment. This data would provide for more accurate analysis of historical manufacturing operations, improve future planning, and leverage other associated processes. Thus, the low-cost vision system developed as a demonstrator was adapted to digitally identify and capture the changes of at least three legacy control panels located on the shopfloor simultaneously. In particular, it was extended with remote access capabilities, an operator's dashboard and local data storage for integration with other production activities. The resulting pilot was retrofitted with a high definition 35 mm camera lens, ruggedized for operation in industrial environments and set up facing the legacy panels using a ceiling mount arm (see Figure 5) at a total cost of under £150. In this way, the camera feeds are processed in real-time to capture changes across the components of the panels, e.g. a light turning on/off. This status change is formatted and presented on the dashboard which is also used for configuration and settings. While this low-cost digital solution meets and addresses the needs of a specific company, it can be quickly repurposed to work with a variety of control panel models found in other manufacturing companies.

Case 2: voice assisted industrial operations
The Hands-Free Digital Drawing Navigation challenge is a form of digital solution where existing equipment is augmented with voice interpretation capabilities so that the end-user can perform operations without physical interaction. The solution developed for this requirement underwent the three stages of the hackathon driven digital solution development process (see Figure 6). In each of these, different levels of ready-made, off-the-shelf, low-cost technologies were put together with the objective of demonstrating proof-of-concept solution ideas, showcase potential benefits and capabilities for manufacturing environments, and enable industrial integration for real-world evaluation.

Stage 1: hackathon solution
The best solution resulting from the Hands-Free Digital Drawing Navigation challenge was a system developed using a USB headset connected to a standard laptop computer equipped with a speech-to-text module, a parsing module, and an application interface module at a total cost of just under £20. In this way, the microphone captures the end-user speech which is fed into the speech-to-text module. The output of this is sent to a parsing module configured to breakdown text into words and select the most syntactically meaningful ones. These are matched against a pre-configured list of commands which are then sent as input to another  computer hosting an application interface module that calls a digital drawing application. In particular, the speech-to-text module was implemented using the open-source, cloud-based Google Speech-to-Text library (see Figure 7) while the other two modules were implemented using Python (i.e. libraries to parse text and libraries to externally interface with Autodesk AutoCAD). While this rapid prototype demonstrates how speech can be used to control a specific computer software, it has the potential to command manufacturing equipment, for example robot arms for delivering material, in situations where the operator is unable to use other human-machine interfaces like dashboards, control panels, or robot controllers.

Stage 2: demonstrator
The development of a solution demonstrator focused on a series of activities designed to extend the scope of the hackathon solution, explore alternative off-the-shelf software solutions for speech-to-text processing as well as to evaluate different options for hardware. Extending the scope of the hackathon solution was focused on demonstrating the use of voice to control industrial equipment, i.e. equipment found on a shopfloor. In order to realize  this, a PLC-controlled turntable with six positions was chosen as representative manufacturing equipment. A microcomputer equipped with a USB microphone and an offline speech-to-text module supporting intent recognition was connected to the PLC via a relay module at a total cost of under £70 (see Figure 8). In this way, when the end-user speaks at the microphone, the voice signal is converted into text which, in turn, is used to identify an intent, i.e. an operation the end-user wants the turntable to perform. This intent is encoded into a binary message and sent as input to the PLC in charge of rotating the turntable to a new position. A variety of open source offline speech-to-text libraries were considered, integrated, and trialed for operations, e.g. Jasper, Rhasspy, and LinTO to name a few. Following this, Rhasspy was chosen because it is widely used across different application domains, hence offering a bigger community for technical support, it is user friendly, it supports industrial standards like MQTT and is in active development. The evaluation of hardware alternatives comprised testing six medium price range microphones with different technologies. In particular, the microphones were systematically compared in terms of performance across a variety of environments including with quiet ambient noise, with ambient noise representative of a factory and with ambient noise representative of a construction site. The microphone tests also considered the end-user speaking at different distances from the microphone. As a result, the demonstrator was equipped with a USB omni-directional condenser microphone featuring noise reduction and echo cancellation.

Stage 3: pilot
One company interested in the solution demonstrator is a manufacturer of inventory vending machines who wanted to expand the capabilities of a specific model by retrofitting voice recognition technology. Technically speaking, it was needed to develop an add-on digital solution that connects to and communicates with a Windows computer controlling both the logic and dispense mechanisms of the vending machine. Hence, the low-cost voice commanded demonstrator was first re-configured with intents associated to the items dispensed by the vending machine. Then, a local network and a REST API were deployed to establish a physical connection and allow communication between the vending machine and the low-cost solution. The resulting pilot was then ruggedized and, together with the microphone and a speaker, physically installed inside the chassis of the vending machine (see Figure 9). All these at a total cost of under £70. Thus, when the enduser requests an item from the machine, the associated intent is sent as input to the Windows computer which, in turn, releases the vending machine latch containing the desired item. As a result of this voice assisted feature, the vending machine offers a reduced user interaction time as it minimizes scrolling through multiple screens.
In addition, such contactless interaction feature enables vending machine operability where there is an inconvenience for end-users to physically interact due to gloves or other type of personal protective equipment. While this low-cost digital solution meets and addresses the needs of a specific company, it can be quickly repurposed to work in different domains, e.g. voice assisted stock control.

Discussion
The hackathon events have revealed that there is a wide variety of readily available hardware and software that can be pulled together to form digital solutions. This is the case for all the presented solutions where participants have not only searched for existing suitable technologies but, more importantly, contributed with their knowledge and expertise. Also, it has been proven that digital solution prototype development can be done quickly and efficiently (if scoped appropriately). In addition, students (i.e. the younger generations) are adept at piecing together the knowledge they need to solve digital problems observed in industry. As an overarching analysis emerging from the events, the results of hackathons are a viable method for businesses, academics, and other organizations to assess whether an idea is worth pursuing further. Furthermore, the quickly delivered lab demonstrators originating from the hackathon events have been beneficial to showcase the potential of digital solutions to manufacturing SMEs. In particular, having these proof-of-concept solutions helped manufacturing organizations realize that it is possible to have simple and cost-efficient alternatives for digitalization. Showcasing multiple solutions gives the chance to assess and compare the actual costs of each implementation in terms of training requirements, ease of integration with existing systems, ways of working, best practice, and other hidden factors which could not be anticipated at the planning phase. In addition, those rapidly made solutions converted into pilots deployed in companies could set a key precedent to demonstrate and explore other application domains where similar type of digital solutions can be adopted after some adaptations.
As it stands, the approach presented here is limited to solution areas emerging from UK manufacturing SMEs. In particular, these organizations have limited financial capabilities for investing in expensive digital solutions and, as a result, have been slow to adopt new digital offerings, therefore making them unable to meet modern requirements and standards of work observed in digitally advanced industries. 4 Considering these characteristics, we believe that the approached presented here holds more widely and could be suitable for the rapid development of low-cost digital solutions of other type of SMEs located in other geographical zones, although more work is needed to validate that. For instance, in an effort to facilitate the identification of affordable digitalization opportunities, the application of the DMS approach has been proposed for logistics SMEs 26 and for construction SMEs. 27

Hackathons for preparing future digital workforce
Today's workforce has limited digital skills and the current digitalization status of UK manufacturing SMEs has shown that workplaces and workforce are unprepared to adapt to Industry 4.0, 28 hence contributing to reduced performance and competitiveness. Since the current adoption pace of digital technologies is slow and future workforce is growing up with domestic digital devices, 29 e.g. mobile and voice operated devices as well as other digital environments, manufacturing organizations, and educational institutions are ought to develop innovative strategies to address the mismatch between the digital skills required in industry and the available skills. Additional ways to enhance educational models have been already proposed, 30 however these changes are usually challenging to implement because they require coordinated support from a number of government, educational, and professional bodies. For this reason, hackathons can be an option to enhance traditional classroom models because of the way they create new and engaging learning experiences and the way they engage students in hands-on practice close to more realistic scenarios. 31 Thus, technical hackathons present students with an opportunity to learn aspects difficult to teach in regular classes since their fast-pace nature could improve and develop technical as well as soft skills beyond academic problems. 32 All of these aspects are evident in the feedback collected at Shoestring hackathons where participants have revealed that they learned a variety of technical and soft skills from a variety of sources difficult to find in regular courses (see Figure 10). Thus, positioning Shoestring hackathons as potential platform for enhancing problem solving and to put digital skills in practice.

Conclusions
The work presented here discussed and reinforced the case for accelerating innovation outputs using cocreation open events, i.e. hackathons. In particular, it presented an overarching approach for accelerating the development of low-cost digital solutions for manufacturing SMEs using Shoestring hackathons. This approach puts together three steps: 1) quickly build hackathon solutions; 2) inspect, test and enhance prototypes into lab demonstrators; and 3) tailor and integrate lab demonstrators into industrial pilots. The application of the approach was demonstrated by revisiting the challenges presented across hackathon events and describing how some of these developments were turned into lab demonstrators before being converted into industrial pilots deployed in manufacturing SMEs.
Using Shoestring hackathons has been beneficial because it gave access to a plethora of ready-made, open source, inexpensive hardware and software technologies that can be integrated to form whole (or parts of) end-to-end digital solutions. In addition, participating teams interpreted low-cost in different ways and solved challenges with different focuses. Thus, allowing multiple approaches from the solution search space to be explored and trialed, including failures and dead-ends, which were not easily predictable a priori. As a result, hackathons helped crystallize requirement uncertainty, i.e. those functional or technical aspects manufacturing SMEs seemed to be unsure about. These outcomes would have been difficult to achieve without an approach capable to ensure accelerated delivery where limited workforce and time represent a challenge.
Defining and employing the hackathon driven digital solution development process was important because it offered a pathway crucial to help accelerate functional prototypes and increase the adoption level of low-cost digitalization. This has been demonstrated by the real time capturing of live operational data and the voice assisted operations solution requirements scoped as hackathon challenges. Both of these solutions started as quick system prototypes developed in hackathon environments, they were subsequently tuned and enhanced in a lab environment for raising low-cost digitalization awareness and, finally, they were tailored and prepared for working within industrial environments. Thus, the proposed approach has been crucial to show manufacturing SMEs that it is possible to introduce digital solutions in very short space of time and that such solutions can be done simple and cheaply. Although the case study associated to the proposed methodology is a suitable approach due to the qualitative nature of this work, this has limitations based on the representativeness of the case study chosen.