Zero-defect manufacturing strategies towards on-line production management for European factories

Zero-defect manufacturing strategies towards on-line production management for European factories

Manufacturing represents approximately 21 % of the EU’s GDP and 20 % of its employment, providing more than 30 million jobs in 230 000 enterprises, mostly SMEs. Moreover, each job in industry is considered to be linked to two more in related services. European manufacturing is also a dominant element in international trade, leading the world in areas such as automotive, machinery and agricultural engineering. Already threatened by both the lower-wage economies and other high-tech rivals, the situation of EU companies was even made more difficult by the downturn.

The Z-Fact0r consortium has conducted an extensive state-of-the-art research and realised that although a number of activities have been trying to address the need for zero-defect manufacturing, still there is a vast business opportunity for innovative, high-ROI (Return on Investment) solutions, that will ensure better quality and higher productivity in the European manufacturing industries.

The Z-Fact0r solution comprises the introduction of five (5) multi-stage production-based strategies targeting (i) the early detection of the defect (Z-DETECT), (ii) the prediction of the defect generation (Z-PREDICT), (iii) the prevention of defect generation by recalibrating the production line (multi-stage), as well as defect propagation in later stages of the production (Z-PREVENT), (iv) the reworking/remanufacturing of the product, if this is possible, using additive and subtractive manufacturing techniques (Z-REPAIR) and (v) the management of the aforementioned strategies through event modelling, KPI (key performance indicators) monitoring and real-time decision support (Z-MANAGE).

To do that we have brought together a total of thirteen (13) EU-based partners, representing both industry and academia, having ample experience in cutting-edge technologies and active presence in the EU manufacturing.

More information
Web resources:
Duration: 42 months
Start date: 01-10-2016
End date: 31-03-2020
Number of participants: 13
Total budget - Public funding: 6 043 019,00 Euro - 4 206 253,00 Euro
Call topic: Zero-defect strategies at system level for multi-stage manufacturing in production lines (FoF.2016.03)
Instrument: Collaborative project (generic)

    Neco use case requirements:

    The following is a list about the most notable requirements NECO has identified and which the new platform developed at the Z-Fact0r project must fulfil:

    • One of the most important requirements is that, in order to pursue an on-line metrology ZDM strategy, the dimensional quality inspection should be performed in 3-4 minutes, which is the average for each machine cycle, due, of course, to the necessity of measuring each and every grinded tap.


    • Automatically measuring the relevant dimensional parameters for each tap comes a close second, because NECO wants to control the stability of the product. In this sense, metrological parameters such as flute length, point core diameter or flute angles should be measured by the new defect detection components.

    The following parameters constitute the complete list of design parameters periodically measured after the grinding operation of taps:

    • Cutting angle: middle of entry
    • Cutting angle: first complete thread position
    • Cutting angle: flute at complete thread position
    • Point core diameter
    • Core diameter at flute
    • Land width at point
    • Land width at flute
    • Flute length

    Further information about them will feature next at the 21, once the tap geometries which have been selected to constitute the Z-Fact0r use cases and defect types have been defined.


    • Checking the feasibility of, and, then, periodically recording some key parameters of the machine, such as vibrations in the wheel axis or grinding wheel speed, so as to have an effective control of the process, as well, and being able to predict machine behaviour and product quality.

    Concerning Microsemi use case: Firstly, the knowledge and physical equipment already in place will have a lasting legacy within the MICROSEMI facility at Caldicot. This understanding and equipment will continue to be used on the original use case and expanded upon as explained in the next section.

    There was a large lack of understanding at the beginning of this project about what could be achieved by the type of approach being undertaken within the project and whether it could produce results in a manufacturing environment. Now we have an answer, and the Z-Fact0r approach has been a great success.

    NECO, as end user, has contributed to the estimation of the benefits Z-Fact0r platform would provide to the manufacturers. In the NECO case:

    • the direct losses for defective product manufacturing at the special flute grinding operation are between 10.000 and 12.000 €/year
    • considering a 25% scrap reduction at this very same operation (technical KPI set for Z-Fact0r, and objectively achievable), some 2.500-3.000€ would be saved yearly in the particular WALTER pilot grinding machine
    • inferring the same defective quantity reduction for all CNC machines due to the implementation of the Z-Fact0r solution, we might be talking about 15.000-20.000€ saved every year (this should be the final estimated figure window for the full implementation of Z-Fact0r at NECO facilities).


    For bringing the Z-Fact0r solution to TRL 9, NECO would definitely support the Consortium intention of trying to find other funding opportunities for the future, especially in the form of other calls from the European Commission.

    For manufacturing of metallic parts with less stringent dimensional requirements, the 3D scanning equipment is a fundamental part of the solution, especially if there are many parts being produced daily with the same geometry. This would allow for an automated inspection to be put in place and, if the production rate is too high, a significative number of samples could still be controlled and used for the PREDICT and PREVENT modules to be used. Of course, sensors in the machines need to be installed for enough data to be available in the machine learning processes. Then, the results could be applied not only to the samples but to all parts being produced. For products with larger added value, the robot deburring is a compulsory tool to use, as it eliminates the need of human intervention in rather sensitive tasks as are the elimination of defects originating in the production line.


        The reduction of the defective rate in the flute grinding process is key, due to this part of the manufacturing chain being the one with the higher proportion of scrapped parts. The time working on the reworking of the parts or the manufacturing of new, “extra” ones, inside a certain batch, multiplied by the hourly cost of the machines would be the numerical estimation of the saved energy, which, of course, is proportional to CO2 emissions. However, the flute grinding process is not the process with the most energy consumption, but it is, by far, the heat treatment (which is located upstream and, thus, the reduction of the defective in this particular process where the NECO pilot focuses would also impact on the energy saving of the machines and processes that come before).


        The solution provided byZ-Fact0r reduces the number of scrapped parts at different steps of the production process. The use of just the needed amount of raw materials is thus a consequence of applying Z-Fact0r, but it has manifold repercussions, as there is reduction of wear of equipment and tools by doing it right the first time. It also eliminates the duplication of auxiliary materials needed to produce the parts.

        The Z-Fact0r platform, and the intensive monitoring of both product and process that accompanies it, has been key for the reduction of defective manufactured parts at the flute grinding process. Should similar solutions be implemented in the rest of machines at NECO premises with a defective rate reduction of around 25%, some 20.000€ could be saved by NECO yearly in the form of raw materials (outside-diameter-grinded High Speed Steel bars), which would instead require to be recycled. Furthermore, if the cutting angle happens to be wrong, the parts cannot be remanufactured and need to be scrapped.


          The Z-Fact0r solution contributes to prevent defective parts to be sintered and reduces the number of defects during the machining of sintered parts. Most of the defective, very hard components, cannot be recovered by further machining or by healing operations and must thus be fully scrapped. Besides these recyclable materials there are also losses on the different production steps: from milling, spray drying, pressing, green machining and sintering.

          With lesser defective part manufacturing, in the same proportion as HSS (raw material), oil consumption from the grinding machine (used as a coolant and in order to reduce friction between the tool and the part) is reduced, and the abrasive material waste generated from the cutting tool is also reduced, something which comes along with the oil as waste waters. Also, further upstream, the usage of salts for the heat treatment process also diminishes.


      Nowadays, it is familiar that within the Industry 4.0, the ICT and the CPS, as parts of the industrial processes, are implemented and merged. For data collection, sensors are being used, imbedded within the AI in order to make smooth communication among humans and machines.  Thus, Z-Fact0r is a pioneer with several advances in predictive maintenance, IoT sensors on shop floors and endless communication between the various components of the system, creating effective and many efficient applications for Industry 4.0.


      Within the Z-Fact0r, the proposed (higher level) DSS, with the support of the knowledge base and the online inspection module (1st level decision support at single stage), produce, verify and validate decisions aligned with the quality control policies, production targets, desired product specifications and maintenance management requirements. Key functional characteristics of the envisioned DSS incorporates among others, techniques for monitoring and predicting product quality, action prioritization, root cause analysis, and mitigation planning algorithms (at product and workstation level). Moving beyond existing solutions that focus only on specific aspects of the production procedure, or that are restrained to diagnosis, the proposed DSS system incorporates autonomous, hierarchical decision support, based on process analytical technologies and newly developed suitably adjusted knowledge-based systems, and  combines product monitoring models and data analytics from heterogeneous sources. The envisioned DSS  takes into account a wide set of multiple factors and criteria, such as data uncertainty, lack of information and information quality, involvement of multiple actors, and real-time response. Thanks to the 5 intertwined zero-defect strategies (i.e. Z-PREDICT, Z-PREVENT, Z-DETECT, Z-REPAIR and Z-MANAGE) the overall solution presents a significant contribution to a spectacular improvement in the overall performance and reliability of the targeted multi-stage manufacturing systems and in the production agility (response to continuous adjustments in production targets).


        DATAPIXEL provides the information associated to the defect detection in the manufacturing parts selected. This information is used as an input for developing the defect detection algorithms of Z-Fact0r solution. Based on this input, a data conditioning methodology has been developed to extract information concerning to the defect position and type. This information will be used as baseline for the model validation, via comparison with the respective simulation results.

        The procedure that has been used is the following:

        • Image-based feature extraction: Convolutional Neural Networks (CNN) and Variational Auto Encoders (VAE) used as feature extractors. CNNs will define the appropriate features that have been used for the classification between healthy and defected parts, and VAEs will be utilized to distinguish that can be used for image generation.
        • Feature selection: An efficient filter feature selection (FS) method was developed for selecting informative and non- redundant feature subsets. In addition to enhanced accuracy rates and dimensionality reduction, the method have reasonably low computational demands. A robust and computationally efficient evaluation criterion with respect to patterns was defined allowing us to assess the redundancy between the features. The proposed FS technique was performed on a forward selection basis handling simultaneously both the discrimination power and the complementary characteristics between the extracted features. To decide on the number of retained features, a termination condition was finally introduced, thus avoiding the trial-and-error procedure usually employed in the common FS techniques of the literature.
        • Classification: The selected features were input in a virtual classification module. The role of this module is to provide a decision on the workpiece condition. The technologies used were Artificial Neural Networks (ANN) and deep learning algorithms.

      Z-Fact0r hybrid framework, obtained by applying a software and hardware integration strategy, is installed on the industrial end users shop floors. This architecture exploits features from Relational Databases and Triplestore while using the blackboard architectural pattern which ensures efficient and accurate communication of data transfer among software applications and devices.


      The overall supervision and optimisation of the system is achieved after the execution of Z-MANAGE strategy. The defects are processed with Decision support system (DSS) tools and are interfaced with Manufacturing Execution Systems (MES). False positives and false negatives are clustered after each Z-Fact0r strategy, which results into a good filtering of these false alarms. To achieve so, the previous acquired knowledge and incidents are also processed to fine tune the system’s operation.

      Additionally, the production is optimised by better scheduling, taking into account the environmental impact of each process. The optimised scheduling and adaptability of the manufacturing improves the overall flexibility, placing a premium on the production rates, satisfying the demand, while preserve increased machinery availability. Since, the Knowledge management system will tune the whole production according to certain quality levels and customer satisfaction, it is highly anticipated that the overall performance of the system will suffice the increased needs of the customers.

      Z-Manage strategy involves also a Knowledge based decision support system which collects knowledge from all the components and the operators and therefore is able to suggest solution for the tuning the rest of the components.

      The strategy involves also the decision making in the event of a defect. The defect will be analysed via the inspection system, from which the defect can be classified and categorised on its severity. In case of “repairable” defects the system will decide for the following; (i) rework on spot, (ii) removal from the production line for further inspection and rework. If the defect is classified as “non-repairable” then the system will decide whether (a) the product will be forwarded to upstream stages, or (b) considered as total failure where it will be recycled.


      Z-DETECT is the first strategy of the Z-Fact0r solution: the detection strategy consists of detecting any machining process anomaly or instability through process monitoring by means of controlled variables called critical process variables (CPVs). In particular, this strategy is invoked when a defect is being generated after the adaptation of the parameters. In such a scenario, an alarm is being triggered to flag the parameters that resulted in a defect. By mapping the true reasons, the system will be able to avoid having more generated defects by weighting the system model.

      Apart from the inspection of the product from which the defect is being observed, the strategy involves more actions and processes to deal both with the generation of the detected defect, and its propagation to the next stages.

      Z-PREDICT strategy is triggered when a defect is recognised during the Z-DETECT stage. The events detected from the physical layer of the system are engineered into high value data that will stipulate new and more accurate process models. Such an unbiased systems behaviour monitoring and analysis provides the basis for enriching the existing knowledge of the system (experience) learning new patterns, raising attention towards behaviour that cause operational and functional discrepancies (e.g. alarms) and the general trends in the shop-floor.

      The more the data pool is being increased the more precise (repeatability) and accurate the predictions will be. The estimations for the future states involve the whole production line, e.g. machine status after x number of operations and/or quality of the products for given set of parameters.

      The system will predict with high confidence the expected quality and customer satisfaction, allowing modifications to the parameters before the production of the products. In addition, Z-Fact0r can operate in the reverse mode, i.e. insert a Customer Satisfaction Goal and control the parameters accordingly to achieve this target.

      The ability of Z-Fact0r to optimise the manufacturing processes according to certain/target quality levels and/or customer satisfaction is the key innovation to fulfil the industrial requirements.


      We are surrounded by services, which are presented to us every day in dozens of different forms; they are so in depth part of our lives that we take for granted their existence, and we get angry when this support is poor of quality or it is not as we would like. Service is certainly not a new topic, although today it has become a topic of discussion in the world of consumers and in the organizations, in the world of politics and economics. In economic terms, the phase in which organisations find themselves is called service revolution, in contrast with the concept of industrial revolution of the twentieth century: the focus has gradually shifted from product to service.

      While at the end of the nineteenth century the major industries and important inventions led to the launching of a large amount of economic goods in the market, now, with the service revolution, we are facing a type of economy based on services. This is due to the shift of attention from the logic of production to customers’ needs and desires, not because industries have begun to think more to individuals than themselves, but for the fact that the production of goods has reached saturation point. Consequently manufacturers have turned their business to new forms to differentiate from competitors in order to survive and be more successful. In a time when raising or lowering the price of goods has no effect on the market, the solution has been found in the possibility of associating a service to the goods, thus providing a value-added to the product.


    Once a “repairable” defect is detected (Z-DETECT), a proper and customized repairing action must be deployed with the minimum time and effort, assuring the best productivity and production flow. In fact, a major challenge for an effective ZD manufacturing is related with the capability to automatically repair the occurred defects without perturbing the overall production flow.

    Z-Fact0r is developing a model-based, supervisory control solution that will be able to interpret the interstage quality control measurements together with the monitoring of the process itself, in order to identify the defect sources and generate a proper and customized repairing action. Additive manufacturing in the form of inkjet or paste printing of various materials (metal, ceramic, polymer resins) can successfully be used to fill a missing spot or correct a damaged part. Upon detection of the defected area, the printing head will deliver the patch material in solution or paste form. In the case of inkjet printing, defect as small as 20 μm can be patched. Post printing treatment of the delivered material include solvent evaporation (e.g. in the case of polymer patches), UV curing (e.g. in the case of epoxy resins) and low temperature laser sintering in the case of metal or ceramic nanoparticles, thermal curable resins or paste where a local reflow process is required.

    To facilitate the implementation of the five strategies, Z-Fact0r is supporting a “reverse supply-chain” policy in the context of a multi-stage supply-chain attached to a multi-stage production. As a result, the defected products/parts detected in downstream stages (produced during a stage, or provided from suppliers in a particular stage) could be returned to upstream stages for remanufacturing or recycling.


      Difficulties in setting up the initial data collection infrastructure on the pilot sites was mitigated through the very careful data collection infrastructure selection where end users have been constantly assisted to set up the best approach to transfer data to Z-Fact0r platform, taking into consideration the internal security policies of the companies.


      Difficulties in having access to machines/systems in the shopfloor for end users’ internal security policies was mitigated with the shopfloor machines/systems interface that is carried out by implementing security measures where different options to avoid direct access to machines through the company network have been investigated (e.g. separate network, mirror DB,…)


      The secure connection are encrypted by two types of authentication:

      • With username and password
      • With X.509 Certificate, the client presents his certificate to the broker during the TLS handshake.

        Z-Fact0r system is a distributed system by desing. The connections between the components and the data flow is authorized by the iLike machine in order the system to maintain all security aspects.



    Data communication between components is essential for the project. End users create data on their shop floor with embedded sensors on the machines, new integrated sensors developed for the project. All these data is propagated in the system with data communication protocols, such as HTTP and AMQP, creating a data stream process in the system. Interoperability between the data communication protocolos and brockers is crucial for a successful result of the data communication of the system. Various data sources work together and use different communication protocols. As a result, all these components and protocols should seamlessly work and their interoperability is what helps them. A message brocker was developed for the project, based on AMQP for data communication. In the initial phases of the project, there were also RESTful APIs that helped in the initial development of the components.


      The whole platform of the Z-Fact0r solution was able to work with other external applications, through a message brocker which is able to receive and send data to external systems. The interoperability level between the Z-Fact0r platform and external applications is essential for communication and integration purposes. Security and safety issues arise when different platforms cooperate. The Z-Fact0r platform implemented an AAA mechanism (Access, Autorisation and Authentication) to secure the safety of the platform during the connection with other external applications.


        Access was given to the Z-Fact0r platform to only authorised users. The platform installation was done either on the shop floor premises or servers deployed by the technical providing partners creating a limited access environment. There was also the authorisation between the components and external appl, where the each component was authorised in an authorisation server with their unique Bearer Token in order to subscribe in the message brocker and publish or receive the available data. Further steps, such as user authentication, were not included in the project scope.


      Z-Fact0r components were developed by different technical providing partners as mostly standalone components. The result was that on each shop floor worked many different components individually. An interoperability level was necessary for the Z-Fact0r system to be a solution to work as a whole. Various integration processes and extensive planning took place during the project and created an integrated system as a final product. The interoperability between the components was the first essential characteristic for the integration process. The components were desinged in the system, in a way that allowed them to operate together without conflicts during data streaming and operation.


        Z-Fact0r architecture was based on the modular design of the components and then the integration of the components to a complete system. For each component a specific architecture was followed by the responsible technology providing partner, base on the use cases, scenarios, end user requirements and technical requirements. The desing for each component was documented in the respective deliverable. Each component also followed the technological trends of their fields and exploited the state of the art of the field. An overall ontology of the Z-Fact0r system was created to include all possible actors, functions, assets etc. All components were initially deployed as standalone applications and then an integration plan was implemented. Z-Fact0r project followed the Incremental Integration Strategy (IIS) where the components were deployed on the shop floors and integrated as one.


        The Z-Fac0r project followed the AMQP and MQTT protocols for the communication between the components. A message brocker was develope by HOLONIX and was called iLike. The iLike implemented the publish/subscribe mechanism for all components which connected to it. The components were authorised in the iLike brocker and repository with a Bearer Token and then used the mechanism to publish or receive the data. An open API was used to create REST GET calls in order to initiate the communicatio between the component and the brocker. The communication steps between the component and the brocker were:

        1. GET call by the component to initiate communication
        2. The component is authenticated by the iLike machine
        3. The iLike machine appoints the respective topics for the components to publish or subscribe to them
        4. A component published data to a topic and another subcribes to the topic to receive the data

        The data from the iLike machines are sent into the cloud to a broker using MQTT protocol (a lightweight protocol that transports messages between devices), it stores the data as messages, so the subscribers can get the values.
        MQTT broker can easily scale from a single device to thousands, manage and tracks all client connection state and permit secure connections.


      The Incremental Integration Strategy (IIS) provides a unified framework for all the EU distributed partners, to work on common principles. By following the IIS, we try to ensure that the integration will be successfully executed in a timely manner. It defines a number of factors to monitor and steps to execute.

      The IIS manifests that the components are integrated and tested incrementally and tested to ensure smooth interaction among them. Every component is combined incrementally, i.e., one by one till all components are integrated logically to make the required application, instead of integrating the whole system at once and then performing testing on the end product. Integrated components are tested as a group to ensure successful integration and data flow between components. The process is repeated until all components are combined and tested successfully. The tests included in the IIS are:

      • Load test
      • Stress test
      • Spike test
      • Endurance test
      • Scalability test
      • Volume test

        Semantic interoperability is desired in the project. An ontology was created to describe all the entities participating in the project components, system, communication protocols as well as the entities given by the end users. The Context Aware algorithm was based on this ontology to create the operation rules for the system. The algorithm provided the essential information to other components about the implementation of the solution. For example, the Context Aware algorithm provided the Reverse Supply Chain with all the necessary information about the production line, the production stages, the return levels and then the RSC was able to create a set of rules to implemented by the end user.


        The data exchange format throughout the project's components was JSON. JSON lightweight, easy for humans to read and write it and provides all relevant information in a formatted way. It is also easy to change to include further fields when necessary or to be restructured for other components. XML was also used as data exchange format. XML also has the same characteristics with JSON in regards to easiness and accessibility. An example of one of the JSON formats used the project is given below to describe the prediction outputs:

        {"holRepoId":3621,"eventDate":1545061853990,"eventDateString":"1545061853990","asset":"Tresky machine","predictedGlueVolume":0.267988,"likelihoodOfDefect":0.0,"defectTypeString":"","insertDate":1583762688099}

        During the integration phase the same communication protocols were used: HTTP and AMQP for the data exchange. Also there is Wi - Fi connection for integration the various components and their updates on premises or on cloud during the integration process of the system. Finally, FTP was used during the integration phase for quick transfer of files on the shop floor premises.


        The IIS and the Integration plan of the Z-Fact0r solution were based on the same APIs and protocols as the data exchange in the system. There were no new APIs designed for the integration process and the integration protocol implemented was derived by the IIS and the Integration plan of the Z-Fact0r system.


    The RESTful API over HTTP has been chosen to fulfil the necessity of sending intermediate or final results to the repository from Z-Modules side, the API utilizes JSON as default exchange format and JWT (JSON Web Token) as authentication mechanisms.
    The JWT is a standard that defines a JSON format scheme for exchanging information between various services. JWTs are widely used to authenticate requests in Web Services authentication mechanisms where the client sends an authentication request to the server, the server generates a signed token and returns it to the client which, from that moment on, will use it to authenticate subsequent requests.


    All database schema, communication protocols, security applications of the Z-Fact0r solution are designed to accommodate the scalability of the solution. All technology can be implemented in larger scale projects without major changes. The one difference with dealing with big data is the use of a different database approach, such as MongoDB, which is more suitable for big data analysis.


    To store data from different sources, including the data elaborated by various Z-Modules a Z-Fact0r data repository has been developed.
    The first source of data is the temporal machine data coming from machine sensors, to store this data is used Cassandra, a distributed NoSQL DBMS capable to handle large amount of data across many servers and provide high availability.
    The following one is used to store others complex and structured production information with the relational DBMS Mysql.
    Another data source in the Z-Fact0r context is the output generated by various modules that carry out the analysis result.