Q&A sessions: Continental
===
###### tags: `AI Proficient` `Use Case` `Continental` `WP1`
Dec 8th, 2020, offline

**RMQ1: Use case will be renames according to standard naming proposed later by Conti**

*RMQ: in the following, by operator we mean the person who is in charge of the job. It may be a jobfloor operator or a decision maker manager*

## General comments:

In each UC we need to have the **baseline** description:
- Is the task already done by the operator or is it a new service to be implemented?
- What is the actual chain of responsabilities?
- What are the present interactions between humans? Are there hierarchical relations? Who is making the decision(s)?
- What are the existing tools?
- Objective metrics: What is the quality obtained today? Speed? Nb of issues per month/year? Time lost?
- Subjective metrics: How does the operator feel about the present process?
- How many operators are involved for each UC and how much do they work together as a group/team?
- Are the operators changing regularly for each UC?
- When a task in the UC is done always by the same operator: how long has the operator been doing that task?
- What are the existng user interfaces like (if any)?
- How much can actions/processes in the UC be traced presently?
- what is the success rate for all UC at the present time?
- what is the minimum success rate threshold to be reached by the service in order to allow it to be deployed?
- What are the cost of errors/scrapping production, etc. for each UC?

Ethical considerations in the use case should be addressed, such as:
- How does the service change the job of the operator?
- How does the service change the relation with its manager? In a direct way and also because AI is in the loop/relation
- Is there some personal data involved?
- Shall data/decisions be traceable?
- When data has been collected already, how long has it been collected for? During the time it was collected how many operators have done the task related to the data?

One should be aware that DL (Deep Learning) has the ability to neglect data that are not usefull to the aim. As such, feature engineering, as usually tackle when bulding more traditional models, should be restricted to the minimum and as many as possible data have to be provided to the DL model for learning. Nevertheless, feature enginerind might be of help when not enough data are available for learning. In the UC all the data involved in are not listed. For instance, in Extruder UC1, the service should help adjust the setting of the combiline but it seems all these settings are not listed (e.g. Temperature, ...).

In the UC descriptions are the 'influenced by' and 'affecting' data/information available?

Almost all UC's involve HMI wich will heavily influence how the operator will feel how service will change his/her work: question of the cognitive load, and perception of the user.

## CONTINENTAL Extrusion UC1: screw feed

**UL foreseen contribution:** Change/anomaly detection could be part of Task 2.2/2.3
What about service to propose settings?

Q1: Data required should identify all the data involved in the UC like extruder temperature and other die area data, data about the "affecting" box....

**Data about the "affecting" box is Existing, but sensors are not placed directly after the extruder, the data is acquired in a different place of the line.**

Q2: what is the time delay to react to the change?
**40 seconds.**

## CONTINENTAL Extrusion UC2: Extrusion restart after >5min stop

**UL contribution:** none

**Propose settings to restart extruder in less time and with less rework while maintaining quality standards.**
Q1: Data required should identify all the data involved in the UC like extruder temperature and other die area data, data about the "affecting" box....
**Existing**
Q2: How is quality checked now?
**quality is checked after production (the extruded product is clearly damaged)**

## CONTINENTAL Extrusion UC3:

In my notes, I wrote "detection" but now I can't see the link. Somebody could provide some insight?
Someone mentioned 'detection problem' in my notes > (detection of deviation in exit conveyor speed or die adjustment?)

Q1: What is an RMEA camera? ‘Retrieval mode electronics assembly’? and what kind of data/images does it take? (Tekniker has similar question)
Q2: Why do the deviations occur: wear of parts or because of operator adjustments?

## CONTINENTAL Conveying UC1: Speed management & breakdown detection

**UL foreseen contribution:** Change/anomaly detection and even prognostic could be part of Task 2.2/2.3

Q1: it is said Conveyor speed (1 point to be added): only one point for hundred meters?
Q2: same question for centering sensor
**yes, one sensor because the environment inside the cooling system is harsh.**
Q3: what are hot/cold delta? does it depends on material and tread shape?

Q4: is there some temperature sensor at the beginning/end of the conveyor?

Q5: does a map of the all available sensor implantation on the combiline exist?

## CONTINENTAL Conveying UC2: Tread cutter

**UL foreseen contribution :** Change/anomaly detection and even prognostic could be part of Task 2.2/2.3

Q1: how does the operator handle this task at the present time?

*The operators check the blade from time to time irregularly. If they notice some wear in the blade because of the cut, they call the maintenance team who comes to check the blade.
The operators are not too interested in checking and they would like a simple alert (a light or something on the main control screen) to show that the blade will need to be changed soon.
The operator can adjust the blade settings a little (angle and speed) to improve the cuts based on different softness of materials. Some operators make better adjustments and some new operators do not bother to adjust much, but this is more or less separate from the blade wear.*

Q2: how do you know that the cutter's blade has to be changed? what is the decision based upon?
*The changing of the blade is decided based upon the product quality further along. If the blade is getting worn (or if the play (=looseness) of bearings or other parts of the arm holding the blade are getting worn or loose) then it begins to leave extra bits in the cut tread, or the cut across the tread is a bit curved, etc. Then it is more difficult to vulcanize the ends of the tread together into a tire later in the process. It was said that it is difficult to infer that the blade is getting work except through the quality of the product.
Paul wants the AI to use the data from new sensors (energy used mostly I think) to predict that the blade will need to be changed in 3 or 4 days, so that curative maintenance can be avoided.*

*There is a 0.2.%/year breakdown rate due to blade wear*

*The idea seems to be that the blade use energy will be collected to establish an average energy, then the AI can predict when the blade will need to be changed. Also if the energy use rises more quickly than it should for only blade wear, e.g. it usually rises to 30% of baseline energy use by a certain number of days, and suddenly it rises to 90% energy use, then they will be alerted that it is probably not the blade wear (only), but some other wear in the blade engine parts, etc.*

Q3: what are 'all' the knife parameters/sensors available?

*The only data now available is how often the blade is changed. They know the day (but not the time) because they write it in the Spare Parts List/Sheet.
Sensors they want to add are included in the Use Case description.
(Aitor) asked if Tekniker can get Continental to do some trial cuttings to get data for the main tread compounds to better understand the energy used and create a baseline; Paul said this was no problem to do.*

## CONTINENTAL Conveying UC3: Blowing station

**UL foreseen contribution :** Change/anomaly detection and even prognostic could be part of Task 2.2/2.3

Q1: how the operator handles this task at the present time?

Q2: how do you know that the tread is perfectly dried (with respect to the available sensors)? what is the decision based upon?

Q3: what are all the blowing station parameters/sensors available?

## CONTINENTAL Tread packaging UC1: Tread alignment

**UL foreseen contribution :** Change/anomaly detection could be part of Task 2.2/2.3

Q1: What does the TBM robot usage KPI mean?
Q2: How many sensors will be needed for conveyor energy?
Q3: What sort of camera/view will be used for the vision on the tread alignment?

**ANY QUESTION?**

## CONTINENTAL Tread packaging UC2: Tread length

**UL foreseen contribution :** Change/anomaly detection could be part of Task 2.2/2.3

Q1: what is the relation/link with 'Conveying UC2: Tread cutter'?
**"Tread cutter" UC is more predictive maintenance, "tread lenght" UC aim to recommend the lenght of the cut based on which product they are manufacturing.**
Q2: how the operator makes decision?

Q3: is a weight sensor available?
**i think : Yes, but not sure where it's placed.**
Q4: in the description 'based on experience' is used does it refer to operator's experience? if not to what experience does it refer?

## CONTINENTAL Tread packaging UC3: Packaging unit use

Q1: what is the baseline now?

Q2: when foreseeing this UC, what could be the consequences for the operator? For instance, when the service will be in use, does the operator has to check if the trolley are well empty in addition to its job? how the operator checks now if the trolley are empty? what will happen if a "tread left on tralley" breakdown happen?

Q3: what are the consequences, now, of the tread left on. the trolley? at the combiline and further on the process?

## CONTINENTAL Quality deviation analysis

**UL foreseen contribution :**

Q1: what is the baseline? how this UC is done now?

Q2: (this question might be of interest in previous UC): usually AI services provides less information than the one used by decision maker... as such decision maker might not be able to handle, in a proper way, the decision/(interaction with service) if he/she doesn't have proper information provided (i.e. the information, he/she, now, based his/her dsecision on)?

Q3: in order to widen the UC, is the pie chart of the 'combiline total time distribution', available in the project description, still relevant:
Shall we have more information about the items: Breakdown, disturbance, preventive maintenance, stopped.

Q4: What is the usual interaction between the Manager doing the analysis and the operators?

Q5: How often is a quality deviation analysis done now?