Suman Acharjee

Domain: Solar Cell & Module Manufacturing Sector ( Capacity: 1.3 GW Solar Cell & 1.6 GW Solar Module Production and Supply per annum) As this has been observed regularly in Solar PV Module Manufacturing, AI is involving increasingly into: In-process Inspection systems (EL imaging, IV curve Analysis, AOI defect detection etc.), SPC monitoring & anomaly alerts, Yield loss Pattern recognition, Predictive maintenance Models, Reliability test analysis the human role shifts significantly. In Solar PV Module Manufacturing as the above mentioned domains are getting AI enabled, hiring should be focusing on following competencies: 1. Root Cause reasoning using AI outputs. 2. Decision Making under uncertain signals. 3. AI Model interpretation and validation. 4. Risk Based thinking instead of following rules. As an example: Interpretation of AI-driven Inspection Result: AI detects the increasing micro cracks in Post Lamination EL inspection with some patterns.A lean practitioner must determine whether the root cause is: Handling Issue? Lamination Pressure Profile imbalance? Cell Supplier change? Hidden stress or vibration from automation? Hence, the hiring process should test the candidate for competency of failure mode reasoning, hypothesis testing. AI detects pattern, Human detects correcting pathway. BORM rejection pattern collaborated with ERP Data: AI is now capable of: Detecting rejection clusters across shifts, machines, operators. Linking rejection trends to specific BORM codes (EVA lot, glass batch, cell supplier, ribbon type) Identifying correlations between material lots and downstream failures Generating automated Pareto and trend dashboards from ERP + MES data But correlation doesn't mean causation. So the executives responsible for improvement must have Data-to-Process translation capability, Multi-layer pattern interpretation (Material, Machine, Operator, Shift) ability, Risk-weighted decision making knowledge, ERP-MES data integrity validation capability, Statistical causality assessment capability. Hence the Hiring must asses: Structured root cause tree building capability. Ability to challenge data before acting Understanding of process-material interaction DOE experience Since this part of ERP System is governed by PPC team, the hiring of PPC team should be evolved also. To govern an AI driven PPC process, a candidate should have competencies of: validating model assumption periodically for optimization of noises. Recalibrating after process changes. Cost-of-poor-quality modeling Supplier negotiation data framing Risk-based prioritization As the Solar PV Module Process and quality part is becoming AI enabled rapidly, the hiring process must be evolved around that accordingly. Without a proper governing knowledge there will be wastage of AI tool resources and chances of catastrophic affects in module processes and unwanted breakdowns in critical machines.

Domain: Solar Cell & Module Manufacturing Sector ( Capacity: 1.3 GW Solar Cell & 1.6 GW Solar Module Production and Supply per annum). When it comes to Responsibility of Decision making, it fundamentally becomes a governance problem. In the domain of manufacturing , Process & Quality control or Solar Engineering, AI primarily works as a support for Decision making , not a decision maker. So the responsibility of Success or Failure both lies around the decision authority, not the support system. Lets start with combination 1: If AI recommendation was followed, but implementation failed. In this case the responsibility of the failure depends on how the governance followed. An MBB must find out, whether AI was validated for this case or not? Were the input data validated? Is the Risk level sufficient for the automated recommendation usage? Were the human review systems utilized properly? If all of these governance are followed properly then responsibility definitely belongs to the Process/System ownership, not the operator who runs the process. This is because the operator followed the approved decision framework and it indicates that there is a limitation in the model or risk assessment was not done properly. In Solar Module Manufacturing, AI can help to conduct DOE or simulations to improve yield of Lamination process. Assume after the analysis AI recommends to increase the temperature to reduce the Bubbles and improve the yield. But after the implementation, the bubbles increases and yield decreases. Now let us check the vastness of the failure: Before the inspection of the lamination process, at peak capacity utilization, there are minimum 120 Solar modules waiting in process and post process. If the governance are followed , due to human review systems for each implementation only 7 or 14 modules will be rejected. So there will be failure but it does not affect the business in large scale. But if the governance was not followed properly then the responsibility shifts towards the human decision maker or the individual operator. If we take the example as above, lets assume that during the hypothesis tests AI indicated low confidence level and the operater ignored that and ran the process without any pilot check. There will be minimum 120 modules rejected. Some manufacturer have 8 or 12 laminators like that. The situation would be catastrophic. Combination 2: If AI recommendation was ignored, but implementation failed. If the deviation was justified, documented and the followings are the causes for the deviation: SME identified contextual factors. safety or compliance are not met. Input data is not reliable. then the responsibility of the failure (or we can say proven wrong) remains with the decision authority. Example: Suppose at Final Testing of modules AI recommends acceptance of marginal module IV curve results, but the quality manager does not allow it after a structured authority discussion, cause it became high risk factor in the last Pre shipment inspection. So he ordered to do frequent calibration of the Sun Simulator and it decreased the capacity utilization. But later it was found that the customer allowed it and there was no need to do so much calibration in the sun-simulator. But again the catastrophe didn't happen. There were less capacity utilization but the lot didn't fail. But what happens if the AI recommended to do the calibration whenever the value of the reference module fails the Nelson rules, but the operator ignores that blindly and it leads to rejection of the lot at PSI stage. This is the true example of AI recommendation was ignored without following the governance. Here the responsibility of the failure solely belongs to the operator and result is also catastrophic. To avoid this confusion, the decision making authority should be structured properly with proper decision responsibility matrix: a. The AI team should only be responsible for Model development and recommendation. b. The responsibility of validation and limits of the model lies with the process governance team. c. Ultimate Operational decisions comes under the authority of the process owner. (In our cases of example: Lamination process owner, Quality manager. Not the lamination operator or the Sun Simulator Operator) d. Management is the owner of Risk level determination before implementation of any autonomous or AI driven changes. A fundamental principle need to be followed without any deviation: AI always recommends- But the process owners are the people who decides. AI should never be a authority, always a highly analytical junior expert.

Domain: Solar Cell & Module Manufacturing Sector ( Capacity: 1.3 GW Solar Cell & 1.6 GW Solar Module Production and Supply per annum). When AI becomes part of a business process, it doesn’t remain static. Inputs change, user behaviour evolves, and the AI’s responses may slowly drift — even if nothing “breaks” outright. Think of a specific process in your domain where AI is/ could be involved in making decisions or recommendations. How would you ensure that the overall process continues to deliver the intended outcomes over time? What would you monitor, and how would you respond when performance starts to slip?Improvement Initiative: Build an AI enabled model for scheduling Preventive Maintenance to improve the following metrics in module manufacturing floor: Pre Lamination FPY%. Cell yield Stringer Uptime. Line Capacity Utilization. Final Quality Yield %. Why is this so important? It has been observed in almost all solar Module Manufacturing lines that the preventive maintenance has been always granted the least priority over continuous running line. This affects the line's Output quality degrading till the start of the PM and the recovery of the line capacity also gets delayed post preventive maintenance hand over. This has led to pre to post PM degradation of : Pre Lamination FPY% by 8-10% Cell yield by 30-50% Stringer Uptime by 20-25% Line Capacity Utilization 20-30% Final Quality Yield % 30 -50%. So instead of improving or maintaining the metrics, PMs often had degraded results for some durations and the business faced losses. These losses made the habit of PM so unpopular in the shop floor that higher management becomes reluctant to activate the PM schedule. If AI can be strengthened with all the historical data and learnings, it can optimize and generate a significant model for Preventive maintenance and the schedule for it. Monitor the Indicators for Positive delivery: Control charts: Though AI claims that the model of the schedule is still going good, the SPC of the above mentioned metrics can tell us the truth. Monthly Management review driven by MBB: As per ISO 9001, AI generated models should be reviewed thoroughly. if the metrics are drifting then the model is definitely failing whether the model accuracy is good or not. Learning correct reality: AI learning reality need to be kept in check. If the engineers acceptance rate is getting lower day by day, it may indicate that the model is not in the correct reality. It can affect catastrophically in near future. Parameter changes: if post PM , the parameters are still inside the recommended window then the model's health is good. Otherwise the model needs to be reverified. Control Strategy when performance slips: Some ground rules need to be confirmed before applying any AI model: There should be one golden PM schedule that is made with experts before starting the implementation, so that if any slip is identified one can revert back to the golden schedule. There should not be any sudden conclusion to scrap the model, first the root cause of incorrect reality needs to be identified and then rectify that. ( Example: Suppose a scheduling of PM for Ribbon Traction area in a stringer within an interval of 2 weeks has degraded the quality of ribbon alignment with cell busbar drastically. So primarily it may be misunderstood that the high frequency of the PM has affected the slot quality of that section and for that the degradation happened. But the reality may be that the frequency of the PM has nothing to do with that. It may be due to: 1. Wrong selection of Technician for such an important section, or 2. Wrong selection of the tools those have been used to do the PM. ) Hence to keep the model healthy and keep away from wrong decisions or reality , a well trained MBB needs to govern the model properly. What makes the model successful: Pro-activeness of triggering PM before the machine's performance starts degrading. Post PM the recovery should be minimized and reduced after every iteration. High positive Business Impact.

Domain: Solar Cell & Module Manufacturing Sector ( Capacity: 1.3 GW Solar Cell & 1.6 GW Solar Module Production and Supply per annum) Improvement Initiative: Improvement of Solar Cell Efficiency from 23% to 25% takes around 6 months by default and it effects the distribution of Module Efficiency positively. But instead of 6 months, the improvement needs to be completed by 4 months to improve the revenue by 30% minimum. DMAIC project can be framed to meet the target as per the initiative. But when AI is introduced — surfacing insights instantly, simulating outcomes, or automating parts of analysis — the sequence, depth, and ownership of DMAIC stages may change. How would you adapt or reinterpret DMAIC in that context? Which stages become stronger with AI, and where must human judgment still dominate? --> DEFINE: Stronger and clearer DEFINE stages make the project reach its target more precisely and within less timeline. AI can help to extract the previous Pareto Charts or analyze the SPC charts faster. It Can help to finalize the CTQ drill down report and find the hidden CTQs (like Improvement of Sheet resistance of the Wafer, Improve the Cell defects during Diffusion / Annealing process. ) quickly. AI Can help to simulate the ROI of the project before the finalization of Project Charter. It can simulate financial impact scenarios in Real time. But still human Judgment is must to: Align all the team members, stakeholders with the target and discuss about their ownership. Differentiate and Prioritization of CTQ which truly matters for customer not just which is statistically viable. To fine tune the AI responses with the help of subject matter experts. ( It can be assumed to have AI as a member of Six Bono Hats Discussions) --> MEASURE: AI has brought a huge breakthrough in case of Measure stage. It can help to: 100% AI driven MSA for measurement devices like EL and AOI systems for defect inspections and Sun Simulators. AI can automatically detect the outliers in the SPC. It can validate the MSA outputs or the detect the faulty human operators by learning the bias during inspections. But human intelligence (HI) still have an upper hand over its artificial counterpart at the time of Selection of 'fit to use' data. Involvement of Regulatory and certification alignments like IEC, BIS, TÜV, UL etc. --> Analysis: AI definitely have more speed and accuracy to analyze data, which helps to: Simulate DOE across thousands of parameters in the machines like Thermal Tools in Solar Cell Parameters. Explore the Multi-dimensional cause effects which human intelligence(HI) can't imagine. identifies the non-linear interactions which conventional six sigma tools can visualize. but to succeed with the above approach HI needs to be used for: check the causality at the shop floor. engineering feasibility for AI recommendations. ethical and safety implication of Parameter change. --> Improve: From the Analysis results AI can easily predict the solutions, optimize the parameters to get to the target easily. Other than that it can also: Simulate multiple pilot runs with different type of scenarios. plan preventive maintenance schedules predict secondary effects.(Increasing diffusion temperature flow rate vs. Cell thermal crack generation) perform multi objective optimization.(Cell Yield, Cell Efficiency, Cost, Warranty Risk) But HI still dominates in : FMEA ownership and sign off. Change management. -->Control: AI plays a strong role while controlling comes into play, it can predict control charts that act before outliers come to the chart. Control the sigma levels continuously. automatically generate corrective action by studying previous actions. Smart audits that target high risk stations proactively. prevents drift instead of reacting to it. Without HI the circle of Control is not complete. The necessity of HI is there when, Governance, Exception handling, Supplier Escalation are needed. Human Intervention is also needed while there is a process change and process owners need to authorize it. Hence all the segments of a DMAIC projects get stronger with the association of AI, but if the project is needed to reach its goal without any unsafe implication or catastrophic results (like 5-10 MW of cells with 80% PID problems while the efficiency is very high, i.e. costly cells) a Human intervention is recommended in all steps. Each steps of DMAIC can be used Fastly with AI algorithms, but a HI is required to ask the right question to AI at right stage, otherwise it can chain react to a false and a very divergent result.

AI-enabled shop floor managers and leaders need to stop involved in the daily matters like Production outputs, Breakdown time, Manpower absenteeism, Raw material availability etc. Instead they have to focus on the matters like: Which process parameters are driving cell-to-module yield loss?-Interaction & Nonlinear Effects, DOE Where is predictive maintenance signaling an impending bottleneck?-Time Series trend analysis, Control charts, Regression, Capability Analysis Which operator or shift pattern correlates with higher defect density?-ANOVA Currently the manufacturing lines are equipped with AI-enabled MES systems which can gather and provide lots of data points that can help the leaders to analyze efficiently. They just need to ask the right questions to the subordinates who are equipped with right tools. Decision making by Generic discussions and previous experience can only delay the process and lead to less efficient results. Leaders need to use AI to store all the good and bad effects of every steps they have taken previously and generate Mistake proofing solutions to avoid failures in the future. During greenbelt projects AI-enables process need to used during FMEA discussions to achieve a faster ramp up target. AI can make all the analysis live on the shop floor, only the mindset of the leaders to implement this on the shop floor needs to be positive and involving.