How do you guarantee accurate data and analytics for manufacturing operations?

Linkedin asked me to provide a perspective since apparently, according to AI, I'm an expert even though I'm not interested in the topic or know anything about it. Anyways, my general perspective is keep doing the things that are working and stop doing the things that aren't working. You're welcome.

The first step is accurate data and that means consistency in instrumentation ( not like that in 737MAX). Without precise and correct data all else is meaningless!

You must start by creating a culture of data literacy throughout the organization. This is forgotten or assumed in too many organizations. Without proper onboarding and continuous training programs, data input, collection, evaluation, and insights are inconsistent at best. Communication across literacy levels is very difficult creating misinformed data "insights" and safety in silos. All leading to the poorest form of "data-driven" decision making. Start with your people.

I agree with this 100%. Establishing a data governance framework is important but equally, having assigned Data Owners within the business is just as important. Without that, Tech will forever be chasing the holy grail. But I also recognise that until you start to analyse the data, we don’t always have a full view of how clean it is and what changes need to be implemented in order to succeed. It has to be a combined effort.

Before even thinking about automation and advanced analytics ask yourself: -What will we do with this data? -Who will use the data? -What will we use it for? Use these as starting criteria to define data collection and governance, start simple, build up from there.

Define clear and understandable criteria for anyone involved in data collection. Implement some validation routines, verify on site that the data being collected is a good reflection of reality.

Data capture must be intergrated within the standard operating procedures. This will ensure consistency and increase reliability of the data recorded. Real time data recording must be carried out to maintain data integrity.

Data quality and normalisation. You must ensure that the data is available for your analysis in the way you expect it. Sometimes a machine maybe down not producing the information you need. Also sometimes machine will output data differently so you need to make sure that all is normalized if you want to calculate a single measurement for all machines like downtime or OEE

Data and most importantly frequency, and calculations governance are key. Also, exceptions or abnormality handling need to be clear in the governance scope.

To guarantee the accurate data in manufacturing operation is not easy. Here my tips : 1. Build your team culture of quality (data not accurate because some thing need to hide). 2. Develop business metrics from top to down and periodically review and validation like : KPI, A3,OKR, act. 3. Develop data colleting with system (dashboard) like : QMS, absenteeism, manufacturing result, act. 4. Analytics data mean the data should be use for value driver for improvement (data is a data, without use the data as andon is waste). 5. Commitment of management to do measurement for business review and performance periodically (by use the number). Thank you

- Identify your key operations management goals. - Look at the types of data you currently collect in your operations. - Consider both descriptive and predictive analytic. - Audit your current data analysis activities. See what types of analysis is currently being done and what tools are used. - Research data analysis methods like statistical process control, root cause analysis....etc - Look into data analysis software like Minitab, JMP, SAS, R, Python, and Tableau that have capabilities to support operations analysis - Consider ease of use and capabilities for visualizing and sharing data insights in addition to analysis power. - Prototype and test different analysis methods and tools

Start with a small scale perhaps representing machine availability, output, scrapped material etc. See that your data is robust enough and then try to use machine learning models to run predictive analytics. Dashboard and othet tools should come into play. But this step relies heavily on the other steps as you need to make sure you can access the data you beed with ease

Further when managing data driven facts, we should take care to exclude outliers objectively. For example if we consider a product development Lead Time that took place during Covid time, it would have been unreal. Supply Chains and normal operational mechanisms got severely disrupted during that time. Similarly if we are looking at Inventory Turns of a parts during market volatility, data may lead to a completely different story to reality. Example would be Inventory Turns of EV battery manufacturer durung supply shortage period. The list goes on for backlog of orders, supply Lead Times, Labour Utilization etc.

I think this is an important step - make sure what you get is not a one time anomaly or totally out thete. Always always communicate your findings at every step of the way. Consult with employees on the plant floor they know best to make sense of the results.

There are various ways to communicate and share manufacturing operations data to stakeholders. In legacy data sharing systems this was communicated using HMI (direct visual inference) through PLC interfaces using data models. This kind of communication was more of descriptive and less dynamic. But in the current digital age, wherein we continually observe disruptions in manufacturing operations, data communication ought to be done in a more real-time based & inferential way. This can be achieved through data models and live dashboards using PowerBI and Python. Such tools are not only used to communicate manufacturing operations analytics in an effective way, but they also can are extremely secure and robust.

Learning and improvement from data analytics from my perspective would largely be model based. Each industry and organization manufactures products wherein the operations process and products follows a specific distribution. Integration of real time data against the statistical distributions will help us learn from current process and enable us to improve by foreseeing tool and machine failures and perform preventive maintenance, improve production yields and also perform root cause analysis by integration through AI/ML.

Consider carefully the design of interfaces used to gather data from humans. In my experience a number of data collection/analytic systems have a soft spot, which is where they require -for example- human operators to key in data. If interface deign, additional workload and ease of access/input are not designed thoughtfully the system achieves poor results in most cases. Two strategies to make the system more robust are: 1. Design out as much as possible human intervention and input requirements choosing to go with sensors & devices to gather data automatically. 2. Design human data input interfaces thoughtfully considering the abilities and motivation of the target workforce.

There is great scope for improvement in process industries, especially those dependent on instrumentation and expert systems.

The use of data analytics is also possible in manufacturing environments that are not automated. This relies on manual entry data capture which is then recorded and summarised on spreadsheets for further analysis. Similar to automated environments, identification of the key metrics to measure is essential. However, there is a delay in the visualisation and communication of resulting data to stakeholders. But it does come though eventually! You cannot improve what you do not measure.