Refine
Year of publication
Institute
- Fachbereich Energietechnik (1116) (remove)
Language
- English (589)
- German (524)
- Multiple languages (1)
- Dutch (1)
- Spanish (1)
Document Type
- Article (604)
- Conference Proceeding (263)
- Book (126)
- Part of a Book (84)
- Doctoral Thesis (10)
- Conference: Meeting Abstract (8)
- Report (7)
- Other (4)
- Talk (3)
- Diploma Thesis (2)
Keywords
- Blitzschutz (18)
- Lightning protection (11)
- Earthquake (5)
- Diversity Management (4)
- Energy storage (4)
- Power plants (4)
- Risikomanagement (4)
- Seismic design (4)
- reinforced concrete (4)
- Associated liquids (3)
Modeling and upscaling of a pilot bayonettube reactor for indirect solar mixed methane reforming
(2020)
A 16.77 kW thermal power bayonet-tube reactor for the mixed reforming of methane using solar energy has been designed and modeled. A test bench for the experimental tests has been installed at the Synlight facility in Juelich, Germany and has just been commissioned. This paper presents the solar-heated reactor design for a combined steam and dry reforming as well as a scaled-up process simulation of a solar reforming plant for methanol production. Solar power towers are capable of providing large amounts of heat to drive high-endothermic reactions, and their integration with thermochemical processes shows a promising future. In the designed bayonet-tube reactor, the conventional burner arrangement for the combustion of natural gas has been substituted by a continuous 930 °C hot air stream, provided by means of a solar heated air receiver, a ceramic thermal storage and an auxiliary firing system. Inside the solar-heated reactor, the heat is transferred by means of convective mechanism mainly; instead of radiation mechanism as typically prevailing in fossil-based industrial reforming processes. A scaled-up solar reforming plant of 50.5 MWth was designed and simulated in Dymola® and AspenPlus®. In comparison to a fossil-based industrial reforming process of the same thermal capacity, a solar reforming plant with thermal storage promises a reduction up to 57 % of annual natural gas consumption in regions with annual DNI-value of 2349 kWh/m2. The benchmark solar reforming plant contributes to a CO2 avoidance of approx. 79 kilotons per year. This facility can produce a nominal output of 734.4 t of synthesis gas and out of this 530 t of methanol a day.
Due to the Renewable Energy Act, in Germany it is planned to increase the amount of renewable energy carriers up to 60%. One of the main problems is the fluctuating supply of wind and solar energy. Here biogas plants provide a solution, because a demand-driven supply is possible. Before running such a plant, it is necessary to simulate and optimize the process. This paper provides a new model of a biogas plant, which is as accurate as the standard ADM1 model. The advantage compared to ADM1 is that it is based on only four parameters compared to 28. Applying this model, an optimization was installed, which allows a demand-driven supply by biogas plants. Finally the results are confirmed by several experiments and measurements with a real test plant.
The seismic performance and safety of major European industrial facilities has a global interest for Europe, its citizens and economy. A potential major disaster at an industrial site could affect several countries, probably far beyond the country where it is located. However, the seismic design and safety assessment of these facilities is practically based on national, often outdated seismic hazard assessment studies, due to many reasons, including the absence of a reliable, commonly developed seismic hazard model for whole Europe. This important gap is no more existing, as the 2020 European Seismic Hazard Model ESHM20 was released in December 2021. In this paper we investigate the expected impact of the adoption of ESHM20 on the seismic demand for industrial facilities, through the comparison of the ESHM20 probabilistic hazard at the sites where industrial facilities are located with the respective national and European regulations. The goal of this preliminary work in the framework of Working Group 13 of the European Association for Earthquake Engineering (EAEE), is to identify potential inadequacies in the design and safety control of existing industrial facilities and to highlight the expected impact of the adoption of the new European Seismic Hazard Model on the design of new industrial facilities and the safety assessment of existing ones.
Globale Stabilitätsanalysen zylindrischer, seismisch belasteter Tanks auf numerischer Grundlage
(2015)
Analysis Of Base Isolated Liquid Storage Tanks With 3D Fsi-Analysis As Well As Simplified Approaches
(2017)
Tanks are preferably designed, for cost-saving reasons, as circular, cylindrical, thin-walled shells. In case of seismic excitation, these constructions are highly vulnerable to stability failures. An earthquake-resistant design of rigidly supported tanks for high seismic loading demands, however, uneconomic wall thicknesses. A cost-effective alternative can be provided by base isolation systems. In this paper, a simplified seismic design procedure for base isolated tanks is introduced, by appropriately modifying the standard mechanical model for flexible, rigidly supported tanks. The non-linear behavior of conventional base isolation systems becomes an integral part of a proposed simplified process, which enables
the assessment of the reduced hydrodynamic forces acting on the tank walls and the corresponding stress distribution. The impulsive and convective actions of the liquid are taken into account. The validity of this approach is evaluated by
employing a non-linear fluid-structure interaction algorithm of finite element method. Special focus is placed on the boundary conditions imposed from the base isolation and the resulting hydrodynamic pressures. Both horizontal and vertical
component of ground motion are considered in order to study the principal effects of the base isolation on the pressure distribution of the tank walls. The induced rocking effects associated with elastomeric bearings are discussed. The results
manifest that base isolated tanks can be designed for seismic loads by means of the proposed procedure with sufficient accuracy, allowing to dispense with numerically expensive techniques.
The article presents the investigation of the seismic behaviour of a modern URM building located in the municipality of Finale Emilia in province of Modena, Northern Italy. The building is situated in the centre of the series of the 2012 Northern Italy earthquakes and has not suffered any damage during the earthquake series in 2012. The observed earthquake resistance of the building is compared with predicted resistances based on linear and nonlinear design approaches according to Eurocode. Furthermore, probabilistic analyses based on nonlinear calculation models taking into account scattering of the most relevant input parameters are carried out to identify their influence to the results and to derive fragility curves.
Eine seismische Anregung verursacht in einem Flüssigkeitstank einen kombinierten Spannungszustand, was zu einem Stabilitätsversagen der häufig sehr dünnwandigen Konstruktionen führen kann. Für die Durchführung von Stabilitätsnachweisen stehen verschiedene Verfahren zur Verfügung. Üblicherweise werden aus Gründen der Einfachheit spannungsbasierte Verfahren angewendet. Diese sind für Einheitslastfälle experimentell abgesichert, wobei eine Übertragung auf kombinierte Spannungszustände wie im Erdbebenfall nur begrenzt möglich ist. Alternativ kann ein globales, numerisches Konzept, das LBA/MNA-Verfahren, angewendet werden. Das Verfahren kombiniert eine materiell nichtlineare Berechnung (MNA) mit einer linearen Beulanalyse (LBA) und erfasst die Interaktion verschiedener gleichzeitig auftretender Beanspruchungen implizit im Nachweis. Dieser Beitrag demonstriert die Anwendung der Verfahren am Beispiel verschiedener Tankgeometrien mit Höhe/Radius-Verhältnissen zwischen 1 ≤ H/R ≤ 2 und Radius/Tankwand-Verhältnissen zwischen 500 ≤ R/t ≤ 1000 und diskutiert zusätzlich die Defizite der spannungsbasierten Nachweisverfahren.
Seismic excited liquid filled tanks are subjected to extreme loading due to hydrodynamic pressures, which can lead to nonlinear stability failure of the thinwalled cylindrical tanks, as it is known from past earthquakes. A significant reduction of the seismically induced loads can be obtained by the application of base isolation systems, which have to be designed carefully with respect to the modified hydrodynamic behaviour of the tank in interaction with the liquid. For this reason a highly sophisticated fluid-structure interaction model has to be applied for a realistic simulation of the overall dynamic system. In the following, such a model is presented and compared with the results of simplified mathematical models for rigidly supported tanks. Finally, it is examined to what extent a simple mechanical model can represent the behaviour of a base isolated tank in case of seismic excitation
The 2012 Emilia-Romagna earthquake, that mainly struck the homonymous Italian region provoking 28 casualties and damage to thousands of structures and infrastructures, is an exceptional source of information to question, investigate, and challenge the validity of seismic fragility functions and loss curves from an empirical standpoint. Among the most recent seismic events taking place in Europe, that of Emilia-Romagna is quite likely one of the best documented, not only in terms of experienced damages, but also for what concerns occurred losses and necessary reconstruction costs. In fact, in order to manage the compensations in a fair way both to citizens and business owners, soon after the seismic sequence, the regional administrative authority started (1) collecting damage and consequence-related data, (2) evaluating information sources and (3) taking care of the cross-checking of various reports. A specific database—so-called Sistema Informativo Gestione Europa (SFINGE)—was devoted to damaged business activities. As a result, 7 years after the seismic events, scientists can rely on a one-of-a-kind, vast and consistent database, containing information about (among other things): (1) buildings’ location and dimensions, (2) occurred structural damages, (3) experienced direct economic losses and (4) related reconstruction costs. The present work is focused on a specific data subset of SFINGE, whose elements are Long-Span-Beam buildings (mostly precast) deployed for business activities in industry, trade or agriculture. With the available set of data, empirical fragility functions, cost and loss ratio curves are elaborated, that may be included within existing Performance Based Earthquake Engineering assessment toolkits.
Monte Carlo Tree Search (MCTS) is a search technique that in the last decade emerged as a major breakthrough for Artificial Intelligence applications regarding board- and video-games. In 2016, AlphaGo, an MCTS-based software agent, outperformed the human world champion of the board game Go. This game was for long considered almost infeasible for machines, due to its immense search space and the need for a long-term strategy. Since this historical success, MCTS is considered as an effective new approach for many other scientific and technical problems. Interestingly, civil structural engineering, as a discipline, offers many tasks whose solution may benefit from intelligent search and in particular from adopting MCTS as a search tool. In this work, we show how MCTS can be adapted to search for suitable solutions of a structural engineering design problem. The problem consists of choosing the load-bearing elements in a reference reinforced concrete structure, so to achieve a set of specific dynamic characteristics. In the paper, we report the results obtained by applying both a plain and a hybrid version of single-agent MCTS. The hybrid approach consists of an integration of both MCTS and classic Genetic Algorithm (GA), the latter also serving as a term of comparison for the results. The study’s outcomes may open new perspectives for the adoption of MCTS as a design tool for civil engineers.
The 2nd edition of the lightning risk management
standard (IEC 62305-2) considers structures, which may
endanger environment. In these cases, the loss is not limited to
the structure itself, which is valid for usual structures. In the past
(Edition 1) this danger was simply taken into account by a special
hazard factor, multiplying the existing risk for the structure with
a number. Now, in the edition 2, we add to the risk for the
structure itself a “second risk” due to the losses outside the
structure. The losses outside can be treated independently from
what occurs inside. This is a major advantage to analyze the risk
for sensitive structures, like chemical plants, nuclear plants, or
structures containing explosives, etc. In this paper, the existing
procedure given by the European version EN 62305-2 Ed.2 is
further developed and applied to a few structures.
In Anbetracht weltweit zunehmend strengerer klimapolitischer
Ziele steigt auch der Druck für Nutzfahrzeughersteller, effizientere und umweltfreundlichere
Technologien zu entwickeln. Den Blick bei der Bewertung dieser
ausschließlich auf die Fahrzeugnutzung zu richten, ist längst nicht mehr zufriedenstellend.
Im Rahmen dieser Analyse wird ein gegenwärtig auf dem Markt erwerblicher
und in deutschen Städten bereits seit Jahren betriebener Hybridbus
energetisch und ökologisch mit einem konventionell angetriebenen, nahezu baugleichen
Modell entlang des Lebensweges bewertet.
Nach Definition von Ziel und Untersuchungsrahmen wird ein Überblick auf bereits
durchgeführte Lebenszyklusanalysen zu Hybridbussen im Stadtverkehr gegeben
und Schlussfolgerungen für die anschließende Analyse abgeleitet. Diese
wird im Rahmen einer energetischen und ökologischen Bewertung beider Produktsysteme
anhand der Parameter "Primärenergieeinsatz" und "CO2äq Emissionen"
praktiziert. Der Fahrzeugrumpf beider Fahrzeuge des gleichen Modells
wird dabei als einheitlich angenommen, sodass bei dem Vergleich der Herstellung
vereinfacht nur die sich unterscheidenden Komponenten des Antriebstranges
berücksichtigt werden. Die Resultate der Wirkungsabschätzung werden als
Differenz des Hybridbusses gegenüber dem Referenzfahrzeug über die einzelnen
Lebenszyklusphasen dargestellt. Schließlich werden Prognosen getroffen, ab
welcher Strecke die bei der Herstellung erzeugten höheren CO2äq Emissionen
des Hybridantriebstranges gegenüber dem Referenzmodell ausgeglichen werden.
In many cities, diesel buses are being replaced by electric buses with the aim of reducing local emissions and thus improving air quality. The protection of the environment and the health of the population is the highest priority of our society. For the transport companies that operate these buses, not only ecological issues but also economic issues are of great importance. Due to the high purchase costs of electric buses compared to conventional buses, operators are forced to use electric vehicles in a targeted manner in order to ensure amortization over the service life of the vehicles. A compromise between ecology and economy must be found in order to both protect the environment and ensure economical operation of the buses.
In this study, we present a new methodology for optimizing the vehicles’ charging time as a function of the parameters CO₂eq emissions and electricity costs. Based on recorded driving profiles in daily bus operation, the energy demands of conventional and electric buses are calculated for the passenger transportation in the city of Aachen in 2017. Different charging scenarios are defined to analyze the influence of the temporal variability of CO₂eq intensity and electricity price on the environmental impact and economy of the bus. For every individual day of a year, charging periods with the lowest and highest costs and emissions are identified and recommendations for daily bus operation are made. To enable both the ecological and economical operation of the bus, the parameters of electricity price and CO₂ are weighted differently, and several charging periods are proposed, taking into account the priorities previously set. A sensitivity analysis is carried out to evaluate the influence of selected parameters and to derive recommendations for improving the ecological and economic balance of the battery-powered electric vehicle.
In all scenarios, the optimization of the charging period results in energy cost savings of a maximum of 13.6% compared to charging at a fixed electricity price. The savings potential of CO₂eq emissions is similar, at 14.9%. From an economic point of view, charging between 2 a.m. and 4 a.m. results in the lowest energy costs on average. The CO₂eq intensity is also low in this period, but midday charging leads to the largest savings in CO₂eq emissions. From a life cycle perspective, the electric bus is not economically competitive with the conventional bus. However, from an ecological point of view, the electric bus saves on average 37.5% CO₂eq emissions over its service life compared to the diesel bus. The reduction potential is maximized if the electric vehicle exclusively consumes electricity from solar and wind power.
This work presents a basic forecast tool for predicting direct normal irradiance (DNI) in hourly resolution, which the Solar-Institut Jülich (SIJ) is developing within a research project. The DNI forecast data shall be used for a parabolic trough collector (PTC) system with a concrete thermal energy storage (C-TES) located at the company KEAN Soft Drinks Ltd in Limassol, Cyprus. On a daily basis, 24-hour DNI prediction data in hourly resolution shall be automatically produced using free or very low-cost weather forecast data as input. The purpose of the DNI forecast tool is to automatically transfer the DNI forecast data on a daily basis to a main control unit (MCU). The MCU automatically makes a smart decision on the operation mode of the PTC system such as steam production mode and/or C-TES charging mode. The DNI forecast tool was evaluated using historical data of measured DNI from an on-site weather station, which was compared to the DNI forecast data. The DNI forecast tool was tested using data from 56 days between January and March 2022, which included days with a strong variation in DNI due to cloud passages. For the evaluation of the DNI forecast reliability, three categories were created and the forecast data was sorted accordingly. The result was that the DNI forecast tool has a reliability of 71.4 % based on the tested days. The result fulfils SIJ’s aim to achieve a reliability of around 70 %, but SIJ aims to still improve the DNI forecast quality.
As part of the transnational research project EDITOR, a parabolic trough collector system (PTC) with concrete thermal energy storage (C-TES) was installed and commissioned in Limassol, Cyprus. The system is located on the premises of the beverage manufacturer KEAN Soft Drinks Ltd. and its function is to supply process steam for the factory's pasteurisation process [1]. Depending on the factory's seasonally varying capacity for beverage production, the solar system delivers between 5 and 25 % of the total steam demand. In combination with the C-TES, the solar plant can supply process steam on demand before sunrise or after sunset. Furthermore, the C-TES compensates the PTC during the day in fluctuating weather conditions. The parabolic trough collector as well as the control and oil handling unit is designed and manufactured by Protarget AG, Germany. The C-TES is designed and produced by CADE Soluciones de Ingeniería, S.L., Spain. In the focus of this paper is the description of the operational experience with the PTC, C-TES and boiler during the commissioning and operation phase. Additionally, innovative optimisation measures are presented.
Large scale central receiver systems typically deploy between thousands to more than a hundred thousand heliostats. During solar operation, each heliostat is aligned individually in such a way that the overall surface normal bisects the angle between the sun’s position and the aim point coordinate on the receiver. Due to various tracking error sources, achieving accurate alignment ≤1 mrad for all the heliostats with respect to the aim points on the receiver without a calibration system can be regarded as unrealistic. Therefore, a calibration system is necessary not only to improve the aiming accuracy for achieving desired flux distributions but also to reduce or eliminate spillage. An overview of current larger-scale central receiver systems (CRS), tracking error sources and the basic requirements of an ideal calibration system is presented. Leading up to the main topic, a description of general and specific terms on the topics heliostat calibration and tracking control clarifies the terminology used in this work. Various figures illustrate the signal flows along various typical components as well as the corresponding monitoring or measuring devices that indicate or measure along the signal (or effect) chain. The numerous calibration systems are described in detail and classified in groups. Two tables allow the juxtaposition of the calibration methods for a better comparison. In an assessment, the advantages and disadvantages of individual calibration methods are presented.
In this work, three patent pending calibration methods for heliostat fields of central receiver systems (CRS) developed by the Solar-Institut Jülich (SIJ) of the FH Aachen University of Applied Sciences are presented. The calibration methods can either operate in a combined mode or in stand-alone mode. The first calibration method, method A, foresees that a camera matrix is placed into the receiver plane where it is subjected to concentrated solar irradiance during a measurement process. The second calibration method, method B, uses an unmanned aerial vehicle (UAV) such as a quadrocopter to automatically fly into the reflected solar irradiance cross-section of one or more heliostats (two variants of method B were tested). The third calibration method, method C, foresees a stereo central camera or multiple stereo cameras installed e.g. on the solar tower whereby the orientations of the heliostats are calculated from the location detection of spherical red markers attached to the heliostats. The most accurate method is method A which has a mean accuracy of 0.17 mrad. The mean accuracy of method B variant 1 is 1.36 mrad and of variant 2 is 1.73 mrad. Method C has a mean accuracy of 15.07 mrad. For method B there is great potential regarding improving the measurement accuracy. For method C the collected data was not sufficient for determining whether or not there is potential for improving the accuracy.
Seismic design of buried pipeline systems for energy and water supply is not only important for plant and operational safety but also for the maintenance of the supply infrastructure after an earthquake. The present paper shows special issues of the seismic wave impacts on buried pipelines, describes calculation methods, proposes approaches and gives calculation examples. This paper regards the effects of transient displacement differences and resulting tensions within the pipeline due to the wave propagation of the earthquake. However, the presented model can also be used to calculate fault rupture induced displacements. Based on a three-dimensional Finite Element Model parameter studies are performed to show the influence of several parameters such as incoming wave angle, wave velocity, backfill height and synthetic displacement time histories. The interaction between the pipeline and the surrounding soil is modeled with non-linear soil springs and the propagating wave is simulated affecting the pipeline punctually, independently in time and space. Special attention is given to long-distance heat pipeline systems. Here, in regular distances expansion bends are arranged to ensure movements of the pipeline due to high temperature. Such expansion bends are usually designed with small bending radii, which during the earthquake lead to high bending stresses in the cross-section of the pipeline. Finally, an interpretation of the results and recommendations are given for the most critical parameters.
Plans for investigations of subthreshold K+ production in p+A collisions / O. W. B. Schult [u.a.]
(1995)
A promising approach to reduce the system costs of molten salt solar receivers is to enable the irradiation of the absorber tubes on both sides. The star design is an innovative receiver design, pursuing this approach. The unconventional design leads to new challenges in controlling the system. This paper presents a control concept for a molten salt receiver system in star design. The control parameters are optimized in a defined test cycle by minimizing a cost function. The control concept is tested in realistic cloud passage scenarios based on real weather data. During these tests, the control system showed no sign of unstable behavior, but to perform sufficiently in every scenario further research and development like integrating Model Predictive Controls (MPCs) need to be done. The presented concept is a starting point to do so.
The integration of high temperature thermal energy storages into existing conventional power plants can help to reduce the CO2 emissions of those plants and lead to lower capital expenditures for building energy storage systems, due to the use of synergy effects [1]. One possibility to implement that, is a molten salt storage system with a powerful power-to-heat unit. This paper presents two possible control concepts for the startup of the charging system of such a facility. The procedures are implemented in a detailed dynamic process model. The performance and safety regarding the film temperatures at heat transmitting surfaces are investigated in the process simulations. To improve the accuracy in predicting the film temperatures, CFD simulations of the electrical heater are carried out and the results are merged with the dynamic model. The results show that both investigated control concepts are safe regarding the temperature limits. The gradient controlled startup performed better than the temperature-controlled startup. Nevertheless, there are several uncertainties that need to be investigated further.
Concerning current efforts to improve operational efficiency and to lower overall costs of concentrating solar power (CSP) plants with prediction-based algorithms, this study investigates the quality and uncertainty of nowcasting data regarding the implications for process predictions. DNI (direct normal irradiation) maps from an all-sky imager-based nowcasting system are applied to a dynamic prediction model coupled with ray tracing. The results underline the need for high-resolution DNI maps in order to predict net yield and receiver outlet temperature realistically. Furthermore, based on a statistical uncertainty analysis, a correlation is developed, which allows for predicting the uncertainty of the net power prediction based on the corresponding DNI forecast uncertainty. However, the study reveals significant prediction errors and the demand for further improvement in the accuracy at which local shadings are forecasted.
Despite the challenges of pioneering molten salt towers (MST), it remains the leading technology in central receiver power plants today, thanks to cost effective storage integration and high cost reduction potential. The limited controllability in volatile solar conditions can cause significant losses, which are difficult to estimate without comprehensive modeling [1]. This paper presents a Methodology to generate predictions of the dynamic behavior of the receiver system as part of an operating assistance system (OAS). Based on this, it delivers proposals if and when to drain and refill the receiver during a cloudy period in order maximize the net yield and quantifies the amount of net electricity gained by this. After prior analysis with a detailed dynamic two-phase model of the entire receiver system, two different reduced modeling approaches where developed and implemented in the OAS. A tailored decision algorithm utilizes both models to deliver the desired predictions efficiently and with appropriate accuracy.
In order to realistically predict and optimize the actual performance of a concentrating solar power (CSP) plant sophisticated simulation models and methods are required. This paper presents a detailed dynamic simulation model for a Molten Salt Solar Tower (MST) system, which is capable of simulating transient operation including detailed startup and shutdown procedures including drainage and refill. For appropriate representation of the transient behavior of the receiver as well as replication of local bulk and surface temperatures a discretized receiver model based on a novel homogeneous two-phase (2P) flow modelling approach is implemented in Modelica Dymola®. This allows for reasonable representation of the very different hydraulic and thermal properties of molten salt versus air as well as the transition between both. This dynamic 2P receiver model is embedded in a comprehensive one-dimensional model of a commercial scale MST system and coupled with a transient receiver flux density distribution from raytracing based heliostat field simulation. This enables for detailed process prediction with reasonable computational effort, while providing data such as local salt film and wall temperatures, realistic control behavior as well as net performance of the overall system. Besides a model description, this paper presents some results of a validation as well as the simulation of a complete startup procedure. Finally, a study on numerical simulation performance and grid dependencies is presented and discussed.
Solares Kochen
(1990)
A new solar desalination system with heat recovery for decentralised drinking water production
(2009)
Entwicklung von Kollektoren und Speicher für solare Prozesswärmesysteme / B. Hafner ; K. Schwarzer
(1996)
We study the possibility to fabricate an arbitrary phase mask in a one-step laser-writing process inside the volume of an optical glass substrate. We derive the phase mask from a Gerchberg–Saxton-type algorithm as an array and create each individual phase shift using a refractive index modification of variable axial length. We realize the variable axial length by superimposing refractive index modifications induced by an ultra-short pulsed laser at different focusing depth. Each single modification is created by applying 1000 pulses with 15 μJ pulse energy at 100 kHz to a fixed spot of 25 μm diameter and the focus is then shifted axially in steps of 10 μm. With several proof-of-principle examples, we show the feasibility of our method. In particular, we identify the induced refractive index change to about a value of Δn=1.5⋅10−3. We also determine our current limitations by calculating the overlap in the form of a scalar product and we discuss possible future improvements.
In collaborative research projects, both researchers and practitioners work together solving business-critical challenges. These projects often deal with ETL processes, in which humans extract information from non-machine-readable documents by hand. AI-based machine learning models can help to solve this problem.
Since machine learning approaches are not deterministic, their quality of output may decrease over time. This fact leads to an overall quality loss of the application which embeds machine learning models. Hence, the software qualities in development and production may differ.
Machine learning models are black boxes. That makes practitioners skeptical and increases the inhibition threshold for early productive use of research prototypes. Continuous monitoring of software quality in production offers an early response capability on quality loss and encourages the use of machine learning approaches. Furthermore, experts have to ensure that they integrate possible new inputs into the model training as quickly as possible.
In this paper, we introduce an architecture pattern with a reference implementation that extends the concept of Metrics Driven Research Collaboration with an automated software quality monitoring in productive use and a possibility to auto-generate new test data coming from processed documents in production.
Through automated monitoring of the software quality and auto-generated test data, this approach ensures that the software quality meets and keeps requested thresholds in productive use, even during further continuous deployment and changing input data.